CN107506738A - Feature extracting method, image-recognizing method, device and electronic equipment - Google Patents

Abstract

A kind of feature extracting method, methods described include：Obtain target image；The target image is divided at least two regions according to the upright direction of personage in the target image；Extract color information of each region in multiple color spaces；The color information of the multiple color spaces extracted from least two region is combined, obtains representing the assemblage characteristic vector of the target image.The present invention also provides a kind of feature deriving means, image-recognizing method and device, electronic equipment.The present invention can extract the feature of image exactly and carry out image recognition and tracking using this feature.

Description

Feature extraction method, image recognition method, device and electronic equipment

technical field

The invention relates to the technical field of image processing, in particular to a feature extraction method, an image recognition method, a device and electronic equipment.

Background technique

With the development of computer technology, there are more and more applications of image processing technology, such as recognizing faces through image recognition technology, tracking people through image tracking technology, etc. Among them, people need to be recognized. However, in character recognition, the posture and movement of the human body often change greatly, so many features cannot be used. Feature extraction is an important technical link of image processing technology. Specifically, feature extraction uses a computer to extract image information to provide a basis for further image processing (such as image matching, image tracking, and person recognition, etc.). If the feature extraction is not complete, sufficient and inaccurate, the information of the image cannot be effectively transmitted, which will affect the subsequent image processing work, such as reducing the accuracy and efficiency of subsequent image recognition and image tracking. Therefore, there is an urgent need for a method for accurately extracting features of an image.

Contents of the invention

In view of this, it is necessary to provide a feature extraction method, image recognition method, device and electronic equipment, which can accurately extract image features and utilize the features for image recognition and tracking.

One aspect of the present invention provides a feature extraction method, the feature extraction method comprising:

Get the target image;

dividing the target image into at least two regions according to the upright direction of the person in the target image;

Extract color information of each region in multiple color spaces;

Combining the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image.

In a possible implementation manner, the extracting the color information of each region in multiple color spaces includes:

According to the color histogram, the information of each color dimension of each region in multiple color spaces is extracted and represented by a feature vector.

In a possible implementation manner, the combining the color information of multiple color spaces extracted from the at least two regions to obtain the combined feature vector representing the target image includes:

Combining the information of each color dimension of each region in each color space to obtain the combined feature vector of each color space in each region;

splicing the combined eigenvectors of each color space in each region to obtain the combined eigenvectors of each region in multiple color spaces;

The combined feature vectors of each area in multiple color spaces are spliced to obtain combined feature vectors of the target image in multiple color spaces of the at least two areas.

In a possible implementation manner, the feature extraction method further includes:

performing dimensionality reduction processing on the combined feature vector representing the target image.

Another aspect of the present invention also provides an image recognition method, the image recognition method comprising:

Get the target image;

dividing the target image into at least two regions according to the upright direction of the person in the target image;

Extract color information of each region in multiple color spaces;

combining the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image;

calculating the distance between the combined feature vector representing the target image and the feature vector of the sample image in the image library to judge the similarity between the target image and the sample image in the image library;

Selecting the sample image with the highest similarity to the target image in the image library as the image recognition result;

The target image is tracked according to the image recognition result.

In a possible implementation manner, the distance between the combined feature vector representing the target image and the feature vector of the sample image in the image library is calculated to determine the distance between the target image and the sample image in the image library. The similarities between previously also include:

performing dimensionality reduction processing on the combined feature vector representing the target image.

Another aspect of the present invention also provides a feature extraction device, which includes:

An acquisition module, configured to acquire a target image;

An area division module, configured to divide the target image into at least two areas according to the upright direction of the person in the target image;

A multi-color space processing module, used to extract color information of each region in multiple color spaces;

A feature expression module, configured to combine the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image.

In a possible implementation manner, the multi-color space processing module is specifically configured to: extract information of each color dimension of each region in multiple color spaces according to the color histogram and represent it with a feature vector.

In a possible implementation manner, the feature expression module is specifically used for:

Combining the information of each color dimension of each region in each color space to obtain the combined feature vector of each color space in each region;

splicing the combined eigenvectors of each color space in each region to obtain the combined eigenvectors of each region in multiple color spaces;

The combined feature vectors of each area in multiple color spaces are spliced to obtain combined feature vectors of the target image in multiple color spaces of the at least two areas.

In a possible implementation manner, the feature extraction device further includes:

A dimensionality reduction module, configured to perform dimensionality reduction processing on the combined feature vector representing the target image.

Another aspect of the present invention also provides an image recognition device, the device comprising:

An acquisition module, configured to acquire a target image;

An area division module, configured to divide the target image into at least two areas according to the upright direction of the person in the target image;

A multi-color space processing module, used to extract color information of each region in multiple color spaces;

A feature expression module, configured to combine the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image;

The image recognition module is used to calculate the distance between the combined feature vector representing the dimensionality reduction of the target image and the feature vector of the sample image in the image library to judge the similarity between the images, and select the image in the image library The sample image with the highest similarity to the target image is an image recognition result, and the target image is tracked according to the image recognition result.

In a possible implementation manner, the image recognition device further includes:

A dimensionality reduction module, configured to perform dimensionality reduction processing on the combined feature vector representing the target image.

Another aspect of the present invention also provides an electronic device, the electronic device includes a memory and a processor, the memory is used to store at least one instruction, and the processor is used to implement the above feature extraction method and method when executing a program stored in the memory. / above image recognition method.

From the above technical solution, in the present invention, the electronic device can acquire the target image; divide the target image into at least two regions according to the direction in which the person stands upright in the target image; extract the color information of each region in multiple color spaces ; Combining the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image. By dividing the target image into regions and extracting color information in the multi-color space of each region, the color of the image is expressed from multiple dimensions, fully reflecting the characteristics and color distribution of different parts of the image, so that no matter what kind of lighting The image under the conditions can accurately represent the image. The purpose of extracting the features of the image is realized accurately through the feature extraction of multiple regions and multiple color spaces.

The present invention adopts the method of dividing the spatial region of the image by the direction of the upright person in the target image, in each region, the color information of multiple color spaces is respectively extracted, the three-dimensional histograms of each color space are spliced into a single dimension, and then connected The histograms of different color spaces in this area, and the histograms connecting all areas to obtain the combined feature vector of multi-area and multi-color spaces, thus overcoming the limitations in the tracking and identification of target objects, increasing the information content of color space features, and at the same time Efficiently represent the spatial information of the target object.

Then reduce the dimensionality of the obtained combined eigenvectors of multi-region and multi-color spaces to reduce the amount of computation and improve the robustness, and then calculate the similarity between images according to the obtained combined eigenvectors of multi-region and multi-color spaces. Identify and track target images.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. Ordinary technicians can also obtain other drawings based on these drawings on the premise of not paying creative work.

Fig. 1 is a flow chart of a feature extraction method provided by an embodiment of the present invention;

Fig. 2 is an example diagram of dividing a target image into at least two regions;

Fig. 3 is a block diagram of a feature extraction device provided by an embodiment of the present invention;

Fig. 4 is a schematic diagram of an electronic device provided by an embodiment of the present invention.

The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings.

detailed description

In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the case of no conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

Many specific details are set forth in the following description to facilitate a full understanding of the present invention, and the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field of the invention. The terms used herein in the description of the present invention are for the purpose of describing specific embodiments only, and are not intended to limit the present invention.

Example

FIG. 1 is a flowchart of a feature extraction method provided by an embodiment of the present invention. Wherein, according to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. As shown in Figure 1, the feature extraction method may include the following steps:

S10: Acquiring a target image.

The above target image is an image to be subjected to feature extraction. For example, if feature extraction is to be performed on an image captured by a camera, the image captured by the camera is used as a target image, and the target image is obtained. Meanwhile, the target image may be an original image captured by the camera, or an image obtained after preprocessing (for example, cutting the original image) on the original image captured by the camera.

The content in the target image may include: people, animals, buildings, scenery, etc., and the content in the target image may include any combination of the above content.

S11: Divide the target image into at least two regions according to the upright direction of the person in the target image.

The proportions of the upright characters are roughly similar, but due to the different postures and actions, the division by the upright direction of the characters will have higher robustness, so the target image is divided into at least two regions according to the upright direction of the characters in the target image.

The above-mentioned upright direction of a person refers to a direction in which a person stands. For example, if the direction in which a person stands in the target image is the Y-axis direction of the Cartesian coordinate system, the target image is divided into at least two regions according to the Y-axis direction of the Cartesian coordinate system. If the standing direction of the person in the target image deviates 45 degrees to the right relative to the Y-axis direction of the Cartesian coordinate system, the target image is divided into at least two regions according to the direction deviated 45 degrees to the right relative to the Y-axis direction of the Cartesian coordinate system.

Please refer to FIG. 2 , which is an example diagram of dividing a target image into at least two regions. Wherein, the upright direction of the person in image 2a is the Y-axis direction of the Cartesian coordinate system, then the image 2a is divided in the Y-axis direction of the Cartesian coordinate system, and the image 2a is represented by a dotted line in the Y-axis direction of the Cartesian coordinate system. 2a is divided into 5 regions.

Meanwhile, in other embodiments, the target image may also be divided into at least two regions according to a direction perpendicular to the upright direction of the person in the image. Please continue to refer to Figure 2. In Figure 2, the upright direction of the person in image 2b is the Y-axis direction of the Cartesian coordinate system, and the direction perpendicular to the upright character in the image is the X-axis direction. Divide, the image 2b is divided into 5 regions in the X-axis direction of the Cartesian coordinate system by dotted lines in the image 2b.

In other embodiments, the upright direction of the person in the target image may also be disregarded, and only be divided according to a preset direction, for example, the image may be divided into regions according to a certain oblique direction. Please continue to refer to Figure 2. Image 2c in Figure 2 is a schematic diagram of dividing the target image in a direction inclined at 45 degrees to the X-axis of the rectangular coordinate system. Divide the image 2c into 5 regions according to the oblique direction.

In other embodiments, the target image is divided into regions according to the spatial pyramid model. How to divide regions according to the spatial pyramid model can be obtained from the prior art, so details will not be described here.

At the same time, when the target image does not include a person, the above method can also be selected to divide the image into regions or other methods can be used to divide the image into regions.

Since in the target image, when divided according to the upright direction of the person, each divided region can contain information about more parts of the person, so dividing according to the upright direction of the person and then extracting features can better reflect the complete image information. For example, when divided according to the upright direction of the character, each area may include multiple parts of the character, such as the head, the upper body of the character, and the lower body of the character; For example, one area includes only the head of the character, another area includes only the upper limbs of the character, and another area includes only the feet of the character.

For example, the upright direction of the characters in image a and image b is the Y-axis direction of the Cartesian coordinate system, image a is a red top and black pants, and image b is a black top and red pants, if it is divided according to the X-axis direction of the Cartesian coordinate system , then when image a and image b are divided and color extraction is performed, the color features extracted from image a and image b are similar, image a and image b will be recognized as the same image, but in fact image a and image b are For two different images, the features extracted at this time are not accurate enough, which further affects the inaccurate recognition results of image recognition and reduces the accuracy of judgment.

Further, when performing region division on the target image, the target image may be divided into 3-5 parts. Dividing too many regions may increase the complexity of the operation and reduce the efficiency of the operation, and dividing too few regions may not be enough to completely distinguish each image. Dividing the target image into 3-5 regions can be enough to distinguish each part of the target image, but not too much. When dividing the area, each area can be an equally divided area or an unequal area.

After the target image is divided into regions, each part can be marked, and each divided region can be identified by marking. For example, the target image is divided into three parts, namely region1, region2 and region3.

If the target image is not divided into regions, it may not be able to fully reflect the spatial orientation characteristics of the object in the target image, thereby reducing the accuracy of image representation. For example, image c is a pedestrian with a red coat and black hat, and image d is a pedestrian with red pants and a black handbag. If only the color information is extracted from image c and image d, the color features extracted from image c and image d are similar, while the actual The above image c and image d are two different images. Therefore, only extracting color information is not enough to represent the complete information of the image, and the image needs to be divided into regions first.

S12: Extract color information of each region in multiple color spaces.

The above color space is also called a color model, and may be any number of color spaces in existing color spaces. Color spaces include but are not limited to RGB color space, CMY color space, YUV color space, HSV color space, HIS color space, and Lab color space.

As an optional implementation manner, in an embodiment, the multiple color spaces may include: an RGB color space, an HSV color space, and a Lab color space.

That is, the multiple color spaces may be three color spaces or more than three color spaces, and if there are three color spaces, they are RGB color space, HSV color space, and Lab color space.

Among them, RGB is the color space adopted in the display, which is defined based on the three primary colors of light. HSV is a color space created according to the intuitive characteristics of color. The parameters in this color space are: hue (H), saturation (S), and brightness (V). Lab is a color space established on the basis of the international standard for color measurement formulated by the International Commission on Illumination (CIE). It is a device-independent color space and a color system based on physiological characteristics. It is described digitally. human visual perception. Therefore, by extracting the color information of RGB color space, HSV color space, and Lab color space, it can be closer to human visual perception, so that the digital color expression is consistent with human perception.

For example, extracting the color information of each area in the RGB color space means: using three primary colors to represent the color information of each area, specifically, a set of R, G, and B values can be used to represent a pixel on each area point. Extract the color information of each region in multiple color spaces, specifically represent the color information of each region in multiple color spaces in digital form, for example, express the color information of each region in multiple color spaces through feature vectors information.

As an optional implementation manner, in an embodiment, the extracting the color information of each region in multiple color spaces includes:

According to the color histogram, the information of each color dimension of each region in multiple color spaces is extracted and represented by a feature vector.

Among them, the color histogram can reflect the composition distribution of colors in the image, that is, which colors appear and the occurrence probability of each color. The above-mentioned color dimensions are also called color channels or color components.

According to the color histogram, the information of each color dimension of each region in multiple color spaces is extracted, specifically, the information of color dimensions is extracted for multiple color spaces in each region, and the extracted colors are represented by feature vectors The color characteristics of the space. The purpose of expressing by eigenvectors is to facilitate subsequent calculations.

When extracting according to the color histogram, first calculate the color histogram of each region. Calculating the color histogram requires dividing the color space into multiple small color intervals, and each small color interval is called a bin of the histogram. This process is called color quantization. The color histogram can then be obtained by counting the number of pixels whose color falls within each small color interval. Generally speaking, the more the number of small color intervals (bins), the stronger the histogram's ability to distinguish colors. However, a color histogram with a large number of bins not only increases the computational burden, but also is not conducive to indexing in large image libraries.

In an embodiment of the present invention, the number of handles in the RGB color space and the HSV color space is set to 8, and the number of handles in the Lab color space is set to 16.

For example, RBG is composed of R, G, and B three dimensions or called three channels, and 8 handles are set for each dimension. HSV consists of three dimensions: H, S, and V, and each dimension also has 8 handles. Lab consists of L (brightness), a (color is from dark green (low brightness value) to gray (medium brightness value) to bright pink (high brightness value)), b (color is from bright blue (low brightness value) ) to gray (medium brightness value) to yellow (high brightness value)) in three dimensions, and each dimension has 16 handles.

If the target image is divided into three regions, region1, region2 and region3, then in a certain region (region1), the information of each color dimension in the extracted RGB color space is expressed as:

[r ₁ , r ₂ , r ₃ , r ₄ , r ₅ , r ₆ , r ₇ , r ₈ ]; [g ₁ , g ₂ , g ₃ , g ₄ , g ₅ , g ₆ , g ₇ , g ₈ ]; [b ₁ , b ₂ , b ₃ , b ₄ , b ₅ , b ₆ , b ₇ , b ₈ ]

Among them, r ₁ is the R value of a pixel in a certain color interval in the color histogram of region1, or the average value of the R values of all pixels in a certain color interval in the color histogram of region1, or The value calculated according to the R value of a preset pixel point in a certain color interval in the color histogram of region1 and the preset formula. Similarly, g ₁ is the G value of a pixel in a certain color interval in the color histogram of region1, or the average value of the G values of all pixels in a certain color interval in the color histogram of region1. Or it is a value calculated according to the G value of a preset pixel point in a certain color interval in the color histogram of region1 and a preset formula.

Similarly, the information of each color dimension in the extracted HSV color space and Lab color space is represented by feature vectors as follows:

[h ₁ , h ₂ , h ₃ , h ₄ , h ₅ , h ₆ , h ₇ , h ₈ ]; [s ₁ , s ₂ , s ₃ , s ₄ , s ₅ , s ₆ , s ₇ , s ₈ ]; [v ₁ , v ₂ , v ₃ , v ₄ , v ₅ , v ₆ , v ₇ , v ₈ ]

[l ₁ ,l ₂ ,l ₃ ,...,l ₁₄ ,l ₁₅ ,l ₁₆ ]; [a ₁ ,a ₂ ,a ₃ ,...,a ₁₄ ,a ₁₅ ,a ₁₆ ];[b ₁ , b ₂ , b ₃ ,..., b ₁₄ , b ₁₅ , b ₁₆ ]

Wherein, the acquisition method of each specific value (such as h ₁ and a ₁ ) in the feature vector can be consistent with the aforementioned method of extracting the information of each color dimension in the RGB color space, or can be based on the method of extracting the information of each color dimension in the RGB color space The result of the information and the color space conversion formula are converted, wherein the color space conversion formula can be obtained from the existing color space conversion formula.

S13: Combine the color information of multiple color spaces extracted from at least two regions to obtain a combined feature vector representing the target image.

After the color information of each area in multiple color spaces is extracted, the color information of multiple areas in multiple color spaces is combined. For example, the color information of each area in multiple color spaces is represented by the feature vector, and then the multiple feature vectors are spliced to obtain the combined feature vector representing the target image, or combined by the mathematical method of the kernel function, A combined feature vector representing the target image is obtained.

As an optional implementation manner, in one embodiment, step S13 combines the color information of multiple color spaces extracted from the at least two regions, and obtaining the combined feature vector representing the target image includes:

Combining the information of each color dimension of each region in each color space to obtain the combined feature vector of each color space in each region;

splicing the combined eigenvectors of each color space in each region to obtain the combined eigenvectors of each region in multiple color spaces;

The combined feature vectors of each area in multiple color spaces are spliced to obtain combined feature vectors of the target image in multiple color spaces of the at least two areas.

Among them, the information of each color dimension of each region in each color space is combined to obtain the combined feature vector of each color space in each region, that is, for the region of region1, it can be expressed that the region is in each color space The eigenvectors of the information of each color dimension of the color are spliced to obtain the eigenvectors of each color space in the region:

[r ₁ , r ₂ ,..., r ₇ , r ₈ , g ₁ , g ₂ ,..., g ₇ , g ₈ , b ₁ , b ₂ ,..., b ₇ , b ₈ ]

[h ₁ , h ₂ ,..., h ₇ , h ₈ , s ₁ , s ₂ ,..., s ₇ , s ₈ , v ₁ , v ₂ ,..., v ₇ , v ₈ ]

[l ₁ , l ₂ ..., l ₁₅ , l ₁₆ , a ₁ , a ₂ , ..., a ₁₅ , a ₁₆ , b ₁ , b ₂ ..., b ₁₆ ]

Then, the combined feature vectors of each color space in each region are spliced, that is, the feature vectors of all color spaces in a region are spliced into a combined feature vector. For the region1, the features representing the information of each color space can be The vectors are spliced to obtain the combined feature vectors of the region in multiple color spaces:

[r ₁ ,...,r ₈ ,g ₁ ,...,g ₈ ,b ₁ ,...,b ₈ ,h ₁ ,...,h ₈ ,s ₁ ,...,s ₈ ,v ₁ ,...,v ₈ ,l ₁ ,...,l ₈ ,a ₁ ,...,a ₈ ,b ₁ ,...,b ₈ ]

=[RGB ₁ HSV ₁ LAB ₁ ]

After the target image is divided into each area, the combined feature vectors in multiple color spaces can be obtained through the above steps, and then the combined feature vectors of each area in multiple color spaces are spliced to obtain the target image in at least two Combined eigenvectors of multiple color spaces for a region:

[Region1, Region2, Region3]

=[RGB ₁ HSV ₁ LAB ₁ , RGB ₂ HSV ₂ LAB ₂ , RGB ₃ HSV ₃ LAB ₃ ]

Through the above steps, the feature extraction of the target image is completed, and the combined feature vector representing the target image is obtained.

As an optional implementation manner, in one embodiment, the feature extraction method provided by the present invention further includes: performing dimensionality reduction processing on the combined feature vector representing the target image.

After obtaining the combined feature vector representing the target image, in order to make the expression more concise and convenient for subsequent use, the combined feature vector representing the target image can be subjected to dimensionality reduction processing to reduce the computational complexity of subsequent use and improve computational efficiency.

For example, the dimensionality reduction method includes principal component analysis (Principal Component Analysis, PCA), which can reduce the dimensionality of the combined feature vector representing the target image through PCA.

At the same time, the method of dimensionality reduction is not limited to the PCA mentioned above, and any dimensionality reduction method such as reverse feature elimination and combination number can also be used to reduce the dimensionality of the combined feature vector representing the target image.

Further, in one embodiment, the combined feature vector of the target image extracted by the feature extraction method provided by the present invention can be used for image recognition and tracking, specifically:

calculating the distance between the combined feature vector representing the target image and the feature vector of the sample image in the image library to judge the similarity between the images;

Selecting the sample image with the highest similarity to the target image in the image library as the image recognition result;

The target image is tracked according to the image recognition result.

Wherein, the combined feature vector representing the target image after dimension reduction or the distance between the combined feature vector representing the target image before dimension reduction and the feature vector of the sample image in the image library can be calculated.

Preferably, calculating the distance between the combined feature vector representing the target image after dimensionality reduction and the feature vector of the sample image in the image library can further improve the efficiency of the operation and reduce the complexity of the operation.

Wherein, the image library may contain information of multiple sample images. For example, the image library includes names, serial numbers, and feature vectors of multiple sample images. For example, it includes the name and serial number of user A’s image, and the feature vector of user A’s image. The sample can be uniquely identified in the image database through the name of user A’s image, and the feature vector of user A’s image can be used to express the characteristics of user A’s image.

When calculating the distance, the distance formula can be used for calculation, and the distance formula includes but not limited to Euclidean distance, Mingren distance, and Mahalanobis distance. Which distance formula to use can be selected according to needs. The distance formula can be obtained from the prior art, and will not be repeated here. There are different methods for determining similarity for different distance formulas. For example, the smaller the distance value obtained according to the Euclidean distance, the greater the similarity between images.

When calculating the distance between the combined feature vector representing the target image and the feature vectors of the sample images in the image library, the combined feature vector representing the target image and the feature vectors of multiple sample images in the image library can be calculated separately to obtain The image with the highest similarity to the target image is used to obtain the image recognition result, that is, whether the target image contains a certain target object.

For example, the M image is subjected to feature extraction to obtain the combined feature vector x representing the M image. If the image library includes A user image, B user image, and C user image, the feature vector of A user image is y _a , and the B user image The feature vector of user C is y _b , and the feature vector of C user image is y _c , then calculate the distances between feature vectors x and y _a , feature vectors x and y _b , and feature vectors x and y _c according to the Euclidean distance. If the calculated distance between the eigenvectors x and y _b is the smallest, the image of user B is identified as the most similar to the target image, and the target image includes user B.

After the recognition result of the target image is obtained, the target image can be tracked according to the image recognition result, specifically, the target object recognized in the target image can be tracked. For example, the B user in the M image is tracked.

When performing image tracking, an existing image tracking algorithm can be selected for tracking, which will not be repeated here.

Since the combined feature vector representing the target image is the result of feature extraction and can be used to represent the target image, the content included in the target image can be identified and tracked according to the result of feature extraction.

If the accuracy of feature extraction of the target image is higher, the accuracy of image recognition and image tracking will be higher; if the accuracy of feature extraction is not high, the accuracy of image recognition and image tracking will also be reduced.

At the same time, in addition to image recognition and image tracking, other image processing operations other than image recognition can also be performed, such as image retrieval. The more accurate the feature extraction of the target image is, the more accurate it will be for subsequent other image processing operations Higher degree and efficiency.

The feature extraction method provided by the embodiment of the present invention acquires a target image; divides the target image into at least two regions according to the upright direction of the person in the target image; extracts color information of each region in multiple color spaces; The color information of multiple color spaces extracted from the at least two regions is combined to obtain a combined feature vector representing the target image. By dividing the target image into regions and extracting color information in the multi-color space of each region, the color of the image is expressed from multiple dimensions, fully reflecting the characteristics and color distribution of different parts of the image, so that no matter what kind of lighting The image under the conditions can accurately represent the image. The purpose of extracting the features of the image is realized accurately through the feature extraction of multiple regions and multiple color spaces.

The present invention adopts the method of dividing the spatial region of the image by the direction of the upright person in the target image, in each region, the color information of multiple color spaces is respectively extracted, the three-dimensional histograms of each color space are spliced into a single dimension, and then connected The histograms of different color spaces in this area, and the histograms connecting all areas to obtain the combined feature vector of multi-area and multi-color spaces, thus overcoming the limitations in the tracking and identification of target objects, increasing the information content of color space features, and at the same time Efficiently represent the spatial information of the target object.

Then reduce the dimensionality of the obtained combined eigenvectors of multi-region and multi-color spaces to reduce the amount of computation and improve the robustness, and then calculate the similarity between images according to the obtained combined eigenvectors of multi-region and multi-color spaces. Identify and track target images.

Example

The present invention also provides an image recognition method, the image recognition method comprising:

Get the target image;

dividing the target image into at least two regions according to the upright direction of the person in the target image;

Extract color information of each region in multiple color spaces;

combining the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image;

calculating the distance between the combined feature vector representing the target image and the feature vector of the sample image in the image library to judge the similarity between the target image and the sample image in the image library;

Selecting the sample image with the highest similarity to the target image in the image library as the image recognition result;

The target image is tracked according to the image recognition result.

As a preferred embodiment, before calculating the distance between the combined feature vector representing the target image and the feature vector of the sample image in the image library, the present invention further includes: Vectors are dimensionally reduced.

After obtaining the combined feature vector representing the target image, in order to make the expression more concise and convenient for subsequent use, the combined feature vector representing the target image can be subjected to dimensionality reduction processing to reduce the computational complexity of subsequent use and improve computational efficiency.

The dimensionality reduction method includes principal component analysis (Principal Component Analysis, PCA), which can reduce the dimensionality of the combined feature vector representing the target image through PCA.

At the same time, the method of dimensionality reduction is not limited to the PCA mentioned above, and any dimensionality reduction method such as reverse feature elimination and combination number can also be used to reduce the dimensionality of the combined feature vector representing the target image.

For the description of the image recognition method, please refer to the description of the relevant steps in the feature extraction method above. I won't repeat them here.

Example

3 is a structural diagram of a feature extraction device provided by an embodiment of the present invention. As shown in FIG.

An acquisition module 201, configured to acquire a target image.

The above target image is an image to be subjected to feature extraction. For example, if feature extraction is to be performed on an image captured by a camera, the image captured by the camera is used as a target image, and the target image is obtained. Meanwhile, the target image may be an original image captured by the camera, or an image obtained after preprocessing (for example, cutting the original image) on the original image captured by the camera.

The content in the target image may include: people, animals, buildings, scenery, etc., and the content in the target image may include any combination of the above content.

The area division module 202 is configured to divide the target image into at least two areas according to the upright direction of the person in the target image.

The proportions of the upright characters are roughly similar, but due to the different postures and actions, the division by the upright direction of the characters will have higher robustness, so the target image is divided into at least two regions according to the upright direction of the characters in the target image.

The above-mentioned upright direction of a person refers to a direction in which a person stands. For example, if the direction in which a person stands in the target image is the Y-axis direction of the Cartesian coordinate system, the target image is divided into at least two regions according to the Y-axis direction of the Cartesian coordinate system. If the standing direction of the person in the target image deviates 45 degrees to the right relative to the Y-axis direction of the Cartesian coordinate system, the target image is divided into at least two regions according to the direction deviated 45 degrees to the right relative to the Y-axis direction of the Cartesian coordinate system.

Please refer to FIG. 2 , which is an example diagram of dividing a target image into at least two regions. Wherein, the upright direction of the person in image 2a is the Y-axis direction of the Cartesian coordinate system, then the image 2a is divided in the Y-axis direction of the Cartesian coordinate system, and the image 2a is represented by a dotted line in the Y-axis direction of the Cartesian coordinate system. 2a is divided into 5 regions.

Meanwhile, in other embodiments, the target image may also be divided into at least two regions according to a direction perpendicular to the upright direction of the person in the image. Please continue to refer to Figure 2. In Figure 2, the upright direction of the person in image 2b is the Y-axis direction of the Cartesian coordinate system, and the direction perpendicular to the upright character in the image is the X-axis direction. Divide, the image 2b is divided into 5 regions in the X-axis direction of the Cartesian coordinate system by dotted lines in the image 2b.

In other embodiments, the upright direction of the person in the target image may also be disregarded, and only be divided according to a preset direction, for example, the image may be divided into regions according to a certain oblique direction. Please continue to refer to Figure 2. Image 2c in Figure 2 is a schematic diagram of dividing the target image in a direction inclined at 45 degrees to the X-axis of the rectangular coordinate system. Divide the image 2c into 5 regions according to the oblique direction.

In other embodiments, the target image is divided into regions according to the spatial pyramid model. How to divide regions according to the spatial pyramid model can be obtained from the prior art, so details will not be described here.

At the same time, when the target image does not include a person, the above method can also be selected to divide the image into regions or other methods can be used to divide the image into regions.

Since in the target image, when divided according to the upright direction of the person, each divided region can contain information about more parts of the person, so dividing according to the upright direction of the person and then extracting features can better reflect the complete image information. For example, when divided according to the upright direction of the character, each area may include multiple parts of the character, such as the head, the upper body of the character, and the lower body of the character; For example, one area includes only the head of the character, another area includes only the upper limbs of the character, and another area includes only the feet of the character.

For example, the upright direction of the characters in image a and image b is the Y-axis direction of the Cartesian coordinate system, image a is a red top and black pants, and image b is a black top and red pants, if it is divided according to the X-axis direction of the Cartesian coordinate system , then when image a and image b are divided and color extraction is performed, the color features extracted from image a and image b are similar, image a and image b will be recognized as the same image, but in fact image a and image b are For two different images, the features extracted at this time are not accurate enough, which further affects the inaccurate recognition results of image recognition and reduces the accuracy of judgment.

Further, when performing region division on the target image, the target image may be divided into 3-5 parts. Dividing too many regions may increase the complexity of the operation and reduce the efficiency of the operation, and dividing too few regions may not be enough to completely distinguish each image. Dividing the target image into 3-5 regions can be enough to distinguish each part of the target image, but not too much. When dividing the area, each area may be an equally divided area or an unequal area.

After the target image is divided into regions, each part can be marked, and each divided region can be identified by marking. For example, the target image is divided into three parts, namely region1, region2 and region3.

If the target image is not divided into regions, it may not be able to fully reflect the spatial orientation characteristics of the object in the target image, thereby reducing the accuracy of image representation. For example, image c is a pedestrian with a red coat and black hat, and image d is a pedestrian with red pants and a black handbag. If only the color information is extracted from image c and image d, the color features extracted from image c and image d are similar, while the actual The above image c and image d are two different images. Therefore, only extracting color information is not enough to represent the complete information of the image, and the image needs to be divided into regions first.

The multi-color space processing module 203 is configured to extract color information of each region in multiple color spaces.

The above color space is also called a color model, and may be any number of color spaces in existing color spaces. Color spaces include but are not limited to RGB color space, CMY color space, YUV color space, HSV color space, HIS color space, and Lab color space.

As an optional implementation manner, in an embodiment, the multiple color spaces may include: an RGB color space, an HSV color space, and a Lab color space.

That is, the multiple color spaces may be three color spaces or more than three color spaces, and if there are three color spaces, they are RGB color space, HSV color space, and Lab color space.

Among them, RGB is the color space adopted in the display, which is defined based on the three primary colors of light. HSV is a color space created according to the intuitive characteristics of color. The parameters in this color space are: hue (H), saturation (S), and brightness (V). Lab is a color space established on the basis of the international standard for color measurement formulated by the International Commission on Illumination (CIE). It is a device-independent color space and a color system based on physiological characteristics. It is described digitally. human visual perception. Therefore, by extracting the color information of RGB color space, HSV color space, and Lab color space, it can be closer to human visual perception, so that the digital color expression is consistent with human perception.

For example, extracting the color information of each area in the RGB color space means: using three primary colors to represent the color information of each area, specifically, a set of R, G, and B values can be used to represent a pixel on each area point. Extract the color information of each region in multiple color spaces, specifically represent the color information of each region in multiple color spaces in digital form, for example, express the color information of each region in multiple color spaces through feature vectors information.

As an optional implementation manner, in one embodiment, the multi-color space processing module 203 is specifically configured to: extract information of each color dimension of each region in multiple color spaces according to the color histogram and represent it with a feature vector.

Among them, the color histogram can reflect the composition distribution of colors in the image, that is, which colors appear and the occurrence probability of each color. The above-mentioned color dimensions are also called color channels or color components.

According to the color histogram, the information of each color dimension of each region in multiple color spaces is extracted, specifically, the information of color dimensions is extracted for multiple color spaces in each region, and the extracted colors are represented by feature vectors The color characteristics of the space. The purpose of expressing by eigenvectors is to facilitate subsequent calculations.

When extracting according to the color histogram, first calculate the color histogram of each region. Calculating the color histogram requires dividing the color space into multiple small color intervals, and each small color interval is called a bin of the histogram. This process is called color quantization. The color histogram can then be obtained by counting the number of pixels whose color falls within each small color interval. Generally speaking, the more the number of small color intervals (bins), the stronger the histogram's ability to distinguish colors. However, a color histogram with a large number of bins not only increases the computational burden, but also is not conducive to indexing in large image libraries.

In an embodiment of the present invention, the number of handles in the RGB color space and the HSV color space is set to 8, and the number of handles in the Lab color space is set to 16.

For example, RBG is composed of R, G, and B three dimensions or called three channels, and 8 handles are set for each dimension. HSV consists of three dimensions: H, S, and V, and each dimension also has 8 handles. Lab consists of L (brightness), a (color is from dark green (low brightness value) to gray (medium brightness value) to bright pink (high brightness value)), b (color is from bright blue (low brightness value) ) to gray (medium brightness value) to yellow (high brightness value)) in three dimensions, and each dimension has 16 handles.

If the target image is divided into three regions, region1, region2 and region3, then in a certain region (region1), the information of each color dimension in the extracted RGB color space is expressed as:

[r ₁ , r ₂ , r ₃ , r ₄ , r ₅ , r ₆ , r ₇ , r ₈ ]; [g ₁ , g ₂ , g ₃ , g ₄ , g ₅ , g ₆ , g ₇ , g ₈ ]; [b ₁ , b ₂ , b ₃ , b ₄ , b ₅ , b ₆ , b ₇ , b ₈ ]

Among them, r ₁ is the R value of a pixel in a certain color interval in the color histogram of region1, or the average value of the R values of all pixels in a certain color interval in the color histogram of region1, or The value calculated according to the R value of a preset pixel point in a certain color interval in the color histogram of region1 and the preset formula. Similarly, g ₁ is the G value of a pixel in a certain color interval in the color histogram of region1, or the average value of the G values of all pixels in a certain color interval in the color histogram of region1. Or it is a value calculated according to the G value of a preset pixel point in a certain color interval in the color histogram of region1 and a preset formula.

Similarly, the information of each color dimension in the extracted HSV color space and Lab color space is represented by feature vectors as follows:

[h ₁ , h ₂ , h ₃ , h ₄ , h ₅ , h ₆ , h ₇ , h ₈ ]; [s ₁ , s ₂ , s ₃ , s ₄ , s ₅ , s ₆ , s ₇ , s ₈ ]; [v ₁ , v ₂ , v ₃ , v ₄ , v ₅ , v ₆ , v ₇ , v ₈ ]

[l ₁ ,l ₂ ,l ₃ ,...,l _14, l ₁₅ ,l ₁₆ ]; [a ₁ ,a ₂ ,a ₃ ,...,a ₁₄ ,a ₁₅ ,a ₁₆ ];[b ₁ , b ₂ , b ₃ ,..., b ₁₄ , b ₁₅ , b ₁₆ ]

Wherein, the acquisition method of each specific value (such as h ₁ and a ₁ ) in the feature vector can be consistent with the aforementioned method of extracting the information of each color dimension in the RGB color space, or can be based on the method of extracting the information of each color dimension in the RGB color space The result of the information and the color space conversion formula are converted, wherein the color space conversion formula can be obtained from the existing color space conversion formula.

The feature expression module 204 is configured to combine the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image.

After the color information of each area in multiple color spaces is extracted, the color information of multiple areas in multiple color spaces is combined. For example, the color information of each area in multiple color spaces is represented by the feature vector, and then the multiple feature vectors are spliced to obtain the combined feature vector representing the target image, or combined by the mathematical method of the kernel function, A combined feature vector representing the target image is obtained.

As an optional implementation manner, in one embodiment, the feature expression module 204 is specifically used to:

Combining the information of each color dimension of each region in each color space to obtain the combined feature vector of each color space in each region;

splicing the combined eigenvectors of each color space in each region to obtain the combined eigenvectors of each region in multiple color spaces;

The combined feature vectors of each area in multiple color spaces are spliced to obtain combined feature vectors of the target image in multiple color spaces of the at least two areas.

Among them, the feature expression module 204 combines the information of each color dimension of each region in each color space to obtain the combined feature vector of each color space in each region, that is, for the region of region1, it can represent that the region is in The eigenvectors of the information of each color dimension of each color space are concatenated to obtain the eigenvectors of each color space in this area:

[r ₁ , r ₂ ,..., r ₇ , r ₈ , g ₁ , g ₂ ,..., g ₇ , g ₈ , b ₁ , b ₂ ,..., b ₇ , b ₈ ]

[h ₁ , h ₂ ,..., h ₇ , h ₈ , s ₁ , s ₂ ,..., s ₇ , s ₈ , v ₁ , v ₂ ,..., v ₇ , v ₈ ]

[l ₁ , l ₂ ,..., l ₁₅ , l ₁₆ , a ₁ , a ₂ ,..., a ₁₅ , a ₁₆ , b ₁ , b ₂ ,..., b ₁₆ ]

Next, the feature expression module 204 splices the combined feature vectors of each color space in each region, that is, splicing the feature vectors of all color spaces in a region into a combined feature vector. For the region region1, each color space can be represented by The eigenvectors of the information are spliced to obtain the combined eigenvectors of the region in multiple color spaces:

[r ₁ ,...,r ₈ ,g ₁ ,...,g ₈ ,b ₁ ,...,b ₈ ,h ₁ ,...,h ₈ ,s ₁ ,...,s ₈ ,v ₁ ,...,v ₈ ,l ₁ ,...,l ₈ ,a ₁ ,...,a ₈ ,b ₁ ,...,b ₈ ]

=[RGB ₁ HSV ₁ LAB ₁ ]

Therefore, each region after the target image is divided can obtain combined feature vectors in multiple color spaces, and the feature expression module 204 splices the combined feature vectors of each region in multiple color spaces to obtain the target image in at least two Combined eigenvectors of multiple color spaces for a region:

[Region1, Region2, Region3]

=[RGB ₁ HSV ₁ LAB ₁ , RGB ₂ HSV ₂ LAB ₂ , RGB ₃ HSV ₃ LAB ₃ ]

The feature extraction of the target image is completed through the feature expression module 204, and a combined feature vector representing the target image is obtained.

As an optional implementation, in one embodiment, the feature extraction device provided by the present invention further includes:

A dimensionality reduction module, configured to perform dimensionality reduction processing on the combined feature vector representing the target image.

After obtaining the combined feature vector representing the target image, in order to make the expression more concise and convenient for subsequent use, the combined feature vector representing the target image can be subjected to dimensionality reduction processing to reduce the computational complexity of subsequent use and improve computational efficiency.

For example, the dimensionality reduction method includes principal component analysis (Principal Component Analysis, PCA), which can reduce the dimensionality of the combined feature vector representing the target image through PCA.

At the same time, the method of dimensionality reduction is not limited to the PCA mentioned above, and any dimensionality reduction method such as reverse feature elimination and combination number can also be used to reduce the dimensionality of the combined feature vector representing the target image.

As an optional implementation, in one embodiment, the combined feature vector of the target image extracted by the feature extraction device provided by the present invention can be used for image recognition and tracking, and when used for image recognition and tracking, can include:

calculating the distance between the combined feature vector representing the target image and the feature vector of the sample image in the image library to judge the similarity between the images;

Selecting the sample image with the highest similarity to the target image in the image library as the image recognition result;

The target image is tracked according to the image recognition result.

Wherein, the combined feature vector representing the target image after dimension reduction or the distance between the combined feature vector representing the target image before dimension reduction and the feature vector of the sample image in the image library can be calculated.

Preferably, calculating the distance between the combined feature vector representing the target image after dimensionality reduction and the feature vector of the sample image in the image library can further improve the efficiency of the operation and reduce the complexity of the operation.

Wherein, the image library may contain information of multiple sample images. For example, the image library includes names, serial numbers, and feature vectors of multiple sample images. For example, it includes the name and serial number of user A's image, and the feature vector of user A's image. The name of user A's image can uniquely identify the sample in the image library, and the feature vector of user A's image can be used to express the characteristics of user A's image.

When calculating the distance, the distance formula can be used for calculation, and the distance formula includes but not limited to Euclidean distance, Mingren distance, and Mahalanobis distance. Which distance formula to use can be selected according to needs. The distance formula can be obtained from the prior art, and will not be repeated here. There are different methods for determining similarity for different distance formulas. For example, the smaller the distance value obtained according to the Euclidean distance, the greater the similarity between images.

When calculating the distance between the combined feature vector representing the target image and the feature vectors of the sample images in the image library, the combined feature vector representing the target image and the feature vectors of multiple sample images in the image library can be calculated separately to obtain The image with the highest similarity to the target image is used to obtain the image recognition result, that is, whether the target image contains a certain target object.

For example, the M image is subjected to feature extraction to obtain the combined feature vector x representing the M image. If the image library includes A user image, B user image, and C user image, the feature vector of A user image is y _a , and the B user image The feature vector of user C is y _b , and the feature vector of C user image is y _c , then calculate the distances between feature vectors x and y _a , feature vectors x and y _b , and feature vectors x and y _c according to the Euclidean distance. If the calculated distance between the eigenvectors x and y _b is the smallest, the image of user B is identified as the most similar to the target image, and the target image includes user B.

After the recognition result of the target image is obtained, the target image can be tracked according to the image recognition result, specifically, the target object recognized in the target image can be tracked. For example, the B user in the M image is tracked.

When performing image tracking, an existing image tracking algorithm can be selected for tracking, which will not be repeated here.

Since the combined feature vector representing the target image is the result of feature extraction and can be used to represent the target image, the content included in the target image can be identified and tracked according to the result of feature extraction.

If the accuracy of feature extraction of the target image is higher, the accuracy of image recognition and image tracking will be higher; if the accuracy of feature extraction is not high, the accuracy of image recognition and image tracking will also be reduced.

At the same time, in addition to image recognition and image tracking, other image processing operations other than image recognition can also be performed, such as image retrieval. The more accurate the feature extraction of the target image is, the more accurate it is for subsequent other image processing operations. Higher degree and efficiency.

In the embodiment of the present invention, the acquisition module acquires the target image; the area division module divides the target image into at least two areas according to the upright direction of the person in the target image; the multi-color space processing module extracts each area in multiple colors Space color information; the feature expression module combines the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image. By dividing the target image into regions and extracting color information in the multi-color space of each region, the color of the image is expressed from multiple dimensions, fully reflecting the characteristics and color distribution of different parts of the image, so that no matter what kind of lighting The image under the conditions can accurately represent the image. The purpose of extracting the features of the image is realized accurately through the feature extraction of multiple regions and multiple color spaces.

The present invention divides the spatial regions of the image according to the upright direction of the person in the target image, extracts color information of multiple color spaces in each region, stitches the three-dimensional histograms of each color space into a single dimension, and then connects the regions The histograms of different color spaces and the histograms of all regions are connected to obtain the combined feature vector of multi-region and multi-color spaces, thereby overcoming the limitations in the tracking and identification of target objects, increasing the amount of information of color space features, and efficiently representing Spatial information of the target object.

Then reduce the dimensionality of the obtained combined eigenvectors of multi-region and multi-color spaces to reduce the amount of computation and improve the robustness, and then calculate the similarity between images according to the obtained combined eigenvectors of multi-region and multi-color spaces. Identify and track target images.

Example

The present invention also provides an image recognition device, which includes:

An acquisition module, configured to acquire a target image;

An area division module, configured to divide the target image into at least two areas according to the upright direction of the person in the target image;

A multi-color space processing module, used to extract color information of each region in multiple color spaces;

A feature expression module, configured to combine the color information of multiple color spaces extracted from the at least two regions to obtain a combined feature vector representing the target image;

The image recognition module is used to calculate the distance between the combined feature vector representing the dimensionality reduction of the target image and the feature vector of the sample image in the image library to judge the similarity between the images, and select the image in the image library The sample image with the highest similarity to the target image is an image recognition result, and the target image is tracked according to the image recognition result.

As a preferred embodiment, the device of the present invention also includes:

A dimensionality reduction module, configured to perform dimensionality reduction processing on the combined feature vector representing the target image.

The dimensionality reduction module may perform dimensionality reduction processing on the combined feature vector representing the target image before calculating the distance between the combined feature vector representing the target image and the feature vectors of the sample images in the image library.

After obtaining the combined feature vector representing the target image, in order to make the expression more concise and convenient for subsequent use, the combined feature vector representing the target image can be subjected to dimensionality reduction processing to reduce the computational complexity of subsequent use and improve computational efficiency.

The dimensionality reduction method includes principal component analysis (Principal Component Analysis, PCA), which can reduce the dimensionality of the combined feature vector representing the target image through PCA.

At the same time, the method of dimensionality reduction is not limited to the PCA mentioned above, and any dimensionality reduction method such as reverse feature elimination and combination number can also be used to reduce the dimensionality of the combined feature vector representing the target image.

For the relevant description of the image recognition device, please refer to the description of the relevant steps in the above-mentioned feature extraction device. I won't repeat them here.

Example

Please refer to FIG. 4 , which is a schematic diagram of an electronic device 1 provided by an embodiment of the present invention. The electronic device 1 includes a memory 20 , a processor 30 and a program 40 stored in the memory 20 and executable on the processor 30 , such as a feature extraction program. When the processor 30 executes the program 40, it realizes the steps in the embodiment of the above feature extraction method, such as steps S10 to S13 shown in FIG. 1; or, when the processor 30 executes the program 40, it realizes the implementation of the above device The functions of each module/unit in the example, such as modules 201-204.

Exemplarily, the program 40 can be divided into one or more modules/units, and the one or more modules/units are stored in the memory 20 and executed by the processor 30 to complete this invention. The one or more modules/units may be a series of program instruction segments capable of accomplishing specific functions, and the instruction segments are used to describe the execution process of the program 40 in the electronic device 1 . For example, the program 40 can be divided into an acquisition module 201, an area division module 202, a multi-color space processing module 203, and a feature expression module 204 in FIG.

The electronic device 1 may be a computer device such as a desktop computer, a notebook computer, a palmtop computer, or a cloud server, or may be an embedded computing device such as a video camera. Those skilled in the art can understand that the schematic diagram 4 is only an example of the electronic device 1, and does not constitute a limitation to the electronic device 1. It may include more or less components than those shown in the figure, or combine certain components, or be different. For example, the electronic device 1 may also include input and output devices, network access devices, buses, and the like.

The so-called processor 30 may be a central processing unit (Central Processing Unit, CPU), and may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor 30 can also be any conventional processor, etc., the processor 30 is the control center of the electronic device 1, and uses various interfaces and lines to connect the entire electronic device 1. various parts.

The memory 20 can be used to store the program 40 and/or module/unit, and the processor 30 runs or executes the computer program and/or module/unit stored in the memory 20, and calls the computer program stored in the memory 20 The data in it realizes various functions of the electronic device 1 . The memory 20 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating device, at least one function required application program (such as a sound playback function, an image playback function, etc.) and the like; the storage data area can be Data created according to use of the electronic device 1 (such as audio data, video data, etc.) and the like are stored. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, flash memory card (Flash Card), at least one magnetic disk storage device, flash memory device, or other volatile solid state storage devices.

If the integrated modules/units of the electronic device 1 are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention realizes all or part of the processes in the methods of the above embodiments, and can also be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. When executed by a processor, the steps in the foregoing method embodiments can be realized. Wherein, the program includes program code, and the program code may be in the form of source code, object code, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, computer-readable media Excludes electrical carrier signals and telecommunication signals.

In the several embodiments provided by the present invention, it should be understood that the disclosed methods and devices may also be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention without limitation. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be Modifications or equivalent replacements can be made without departing from the spirit and scope of the technical solutions of the present invention.

Claims (11)

1. a kind of feature extracting method, it is characterised in that methods described includes：

Obtain target image；

The target image is divided at least two regions according to the upright direction of personage in the target image；

Extract color information of each region in multiple color spaces；

The color information of the multiple color spaces extracted from least two region is combined, obtains representing the target The assemblage characteristic vector of image.

2. the method as described in claim 1, it is characterised in that believe in the color of multiple color spaces in each region of extraction Breath includes：

Each region is extracted in multiple color spaces according to color histogram and the information of each color dimension and uses characteristic vector It is indicated.

3. method as claimed in claim 2, it is characterised in that it is described will be extracted from least two region it is multiple The color information combination of color space, obtains representing that the assemblage characteristic vector of the target image includes：

By each region in the information combination of each color dimension of each color space, it is empty to obtain each color in each region Between assemblage characteristic vector；

The assemblage characteristic vector of each color space in each region is spliced, obtains each region in multiple colors The assemblage characteristic vector in space；

Each region is spliced in the assemblage characteristic vector of multiple color spaces, obtains the target image described The assemblage characteristic vector of multiple color spaces at least two regions.

4. the method as described in claim 1, it is characterised in that methods described also includes：

Dimension-reduction treatment is carried out to the assemblage characteristic vector for representing the target image.

5. a kind of image-recognizing method, it is characterised in that methods described includes：

Obtain target image；

The target image is divided at least two regions according to the upright direction of personage in the target image；

Extract color information of each region in multiple color spaces；

The color information of the multiple color spaces extracted from least two region is combined, obtains representing the target The assemblage characteristic vector of image；

Calculate in the assemblage characteristic vector for representing the target image and image library between the characteristic vector of sample image Distance is to judge the similitude in target image and described image storehouse between sample image；

The similitude highest sample image chosen in described image storehouse with the target image is image recognition result；

The target image is tracked according to described image recognition result.

6. image-recognizing method as claimed in claim 5, it is characterised in that calculate the expression target image described Assemblage characteristic vector and sample image in image library the distance between characteristic vector to judge target image and described image Also include before similitude in storehouse between sample image：

Dimension-reduction treatment is carried out to the assemblage characteristic vector for representing the target image.

7. a kind of feature deriving means, it is characterised in that described device includes：

Acquisition module, for obtaining target image；

Region division module, for being divided into the target image at least according to the upright direction of personage in the target image Two regions；

Multiple color spaces processing module, for extracting color information of each region in multiple color spaces；

Feature representation module, for the color information group for the multiple color spaces that will be extracted from least two region Close, obtain representing the assemblage characteristic vector of the target image.

8. feature deriving means as claimed in claim 7, it is characterised in that described device also includes：

Dimensionality reduction module, for carrying out dimension-reduction treatment to the assemblage characteristic vector for representing the target image.

9. a kind of pattern recognition device, it is characterised in that described device includes：

Acquisition module, for obtaining target image；

Region division module, for being divided into the target image at least according to the upright direction of personage in the target image Two regions；

Multiple color spaces processing module, for extracting color information of each region in multiple color spaces；

Feature representation module, for the color information group for the multiple color spaces that will be extracted from least two region Close, obtain representing the assemblage characteristic vector of the target image；

Picture recognition module, for calculate the assemblage characteristic vector after the dimensionality reduction for representing the target image with image library The distance between characteristic vector of sample image to judge the similitude between image, choose in described image storehouse with the target The similitude highest sample image of image is image recognition result, and the target image is entered according to described image recognition result Line trace.

10. device as claimed in claim 9, it is characterised in that described device also includes：

Dimensionality reduction module, for representing that the assemblage characteristic vector of the target image carries out dimension-reduction treatment.

11. a kind of electronic equipment, it is characterised in that the electronic equipment includes memory and processor, and the memory is used for At least one instruction is stored, the processor is realized when being used to perform the program stored in memory as appointed in claim 1-4 One feature extracting method of meaning and/or such as claim 5-6 described image recognition methods.