CN111598177A - An Adaptive Maximum Sliding Window Matching Method for Low Overlap Image Matching - Google Patents

Abstract

An adaptive maximum sliding window matching method, comprising the following steps: dividing an image to be matched into a reference image and a sliding image, setting an estimated matching position (x ₀ , y ₀ ) between the images to be matched, and setting the number of horizontal search pixels to R _x and the number of vertical search pixels is R _y ; calculate the horizontal sliding range T _x and vertical sliding range T _y of the sliding image; move the sliding image pixel by pixel within the sliding range, and calculate the similarity of the overlapping area; record the maximum overlapping sliding window similarity The highest position is the best matching position of the image. The method is not limited by the size of the images to be matched and the size of the overlapping area, and is especially suitable for automatic matching processing between low-overlap images. And by adaptively adjusting the sliding window, the similarity of the maximum overlapping area between the images to be matched is measured, and more image information is used to improve the accuracy of the matching result.

Description

An Adaptive Maximum Sliding Window Matching Method for Low Overlap Image Matching

technical field

The invention relates to the technical field of image matching, in particular to an adaptive maximum sliding window matching method for low-overlap image matching.

Background technique

Image matching is an important image processing method to realize the comprehensive application of multiple image information, and has a wide range of applications in computer vision, remote sensing data processing and other fields. Existing image matching methods mainly include feature-based methods and region-based methods. The SIFT feature matching method is a classic feature-based image matching method. The matching process mainly includes building a scale space and detecting significant feature points, generating feature descriptors, and performing feature similarity measurement to obtain image matching relationships. The SIFT feature scale space construction process and feature descriptor production process are shown in Figure 1(a) and Figure 1(b), respectively.

Template matching method is a commonly used region-based image matching method. The template matching method divides the image to be matched into a reference image and a template image. The template image is translated within the range of the reference image, and the similarity between the template image and the reference image covered by it is calculated to obtain the best matching position. The schematic diagram of the processing is shown in the figure. 2, where f represents the reference image to be matched, t represents the template image to be matched, assuming that the width of the reference image is w, the height is h, the width of the template image is w ₀ , and the height is h ₀ , then the template matches There is a certain search range limitation, that is, the horizontal search range is 0 to w-w0, and the vertical search range is 0 to hh ₀ .

The common point of image matching methods is that there is a certain overlapping area between the images to be matched. The feature detection and feature description process based on the SIFT feature matching method has obvious requirements on the size and texture saliency of the common area of the image to be matched. In the method based on template matching, the template image needs to be slid within the range of the reference image, the overlapping area between the images to be imaged is required to be large, and the search range is limited. For the case where the overlapping area between the images to be matched is small, such as between the CCD imaging slices of the optical remote sensing camera, the strip images obtained by the optical remote sensing camera swing imaging mode, etc., the existing matching methods cannot use the overlapping information between the images to the maximum extent. , therefore, accurate and stable matching results cannot be obtained.

In the process of realizing the present invention, the applicant found that the above-mentioned existing single remote sensing image target height calculation technology has the following technical defects:

(1) The existing matching methods need to select point features or templates of a certain scale, which limits the minimum size of the overlapping area of the images to be matched, and therefore cannot be applied to matching processing between images with narrow overlapping areas;

(2) The existing matching methods do not fully use the information of the overlapping area, and only perform image matching by measuring the information around the feature or the template area, and the matching stability is insufficient.

SUMMARY OF THE INVENTION

In view of this, the main purpose of the present invention is to provide an adaptive maximum sliding window matching method for low-overlap image matching, so as to partially solve at least one of the above technical problems.

In order to achieve the above object, as an aspect of the present invention, an adaptive maximum sliding window matching method for low-overlap image matching is provided, comprising the following steps:

Divide the image to be matched into a reference image and a sliding image, set the estimated matching position (x ₀ , y ₀ ) between the images to be matched, and set the number of horizontal search pixels to be R _x and the number of vertical search pixels to be R _y ;

Calculate the horizontal sliding range T _x and the vertical sliding range _Ty of the sliding image;

Move the sliding image pixel by pixel within the sliding range, and calculate the similarity of the overlapping area;

Record the position with the highest similarity of the maximum overlapping sliding window, which is the best matching position of the image.

Wherein, the horizontal sliding range T _x and the vertical sliding range T _y of the sliding image are:

Wherein, the specific calculation process for calculating the similarity of overlapping regions is as follows:

Set the moving position of the sliding image in the search range as (x, y), calculate the maximum overlapping area range at this time, and obtain the maximum overlapping area g _f of the reference image and the overlapping area g _t of the sliding image;

Calculate the similarity of the corresponding overlapping regions g _f and _gt .

Wherein, the calculation formula of the upper left corner coordinates (x _f , y _f ) of the overlapping area of the reference image is as follows:

The formula for calculating the coordinates (x′ _f , y′ _f ) of the lower right corner of the overlapping area of the reference image is as follows:

Wherein, the calculation formula of the upper left corner coordinates (x _t , y _t ) of the overlapping area of the sliding image is as follows:

The formula for calculating the coordinates (x' _t , y' _t ) of the lower right corner of the overlapping area of the sliding image is as follows:

Wherein, the calculation of the similarity of the corresponding overlapping regions g _f and g _t adopts a normalized cross-correlation similarity measurement method, standard square deviation or standard correlation matching.

Wherein, the normalized cross-correlation coefficient is used to measure the similarity, and the specific calculation formula is as follows:

Among them, (u, v) is the coordinate grid position of the g _f and g _t images, which is calculated pixel by pixel, and E(g _f ) and E(g _t ) are the gray mean values of the images g _f and g _t , respectively.

Based on the above technical solutions, the adaptive maximum sliding window matching method for low-overlap image matching of the present invention has at least one of the following beneficial effects relative to the prior art:

(1) The present invention proposes an adaptive maximum sliding window matching method, which performs image matching by calculating the similarity of the maximum overlapping area between the images to be matched. Therefore, it is not limited by the size of the image to be matched and the size of the overlapping area, and is especially suitable for low overlap. Automatic matching processing between degree images.

(2) By adaptively adjusting the sliding window, the similarity of the maximum overlapping area between the images to be matched is measured, and more image information is used to improve the accuracy of the matching result.

Description of drawings

Fig. 1(a) is the process of constructing a Gaussian scale space by the SIFT feature matching method in the prior art, and Fig. 1(b) is a schematic diagram of the generation process of the SIFT feature descriptor in the prior art;

2 is a schematic diagram of a template matching method in the prior art;

3 is a schematic diagram of a template matching method of the present invention;

4 is a flowchart of an adaptive maximum sliding window matching method for low-overlapping image matching of the present invention;

5 is a schematic diagram of low-overlap matching data according to an embodiment of the present invention;

6 is a schematic diagram of the best matching position obtained by adaptive maximum sliding window matching according to an embodiment of the present invention;

FIG. 7 is a schematic diagram after splicing and displaying two images based on a matching result according to an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

Aiming at the determination of the prior art solution, the present invention discloses an adaptive maximum sliding window matching method for low overlapping image matching. Not limited by the size of the overlapping area, it can be applied to matching between low-overlapping images, and in extreme cases, it can be applied to matching between images with only a single-column overlapping area; 2) By adaptively adjusting the maximum sliding window, the image information of the overlapping area is fully used, Unaffected by the distribution of salient features of the image, the matching results are more accurate and stable. The technical scheme of the present invention is as follows:

1. Divide the image to be matched into a reference image f and a sliding image t, assuming that the width of the reference image is w, the height is h, the width of the sliding image is w ₀ , and the height is h ₀ , as shown in Figure 3;

2. Calculate the horizontal sliding range T _x and the vertical sliding range _Ty of the sliding image. Assuming that the estimated matching position between the images to be matched is (x ₀ , y ₀ ), the number of horizontal search pixels is R _x and the number of vertical search pixels is R _y , the sliding range of the sliding image is:

T _x ∈[x ₀ -R _x , x ₀ +R _x ]

T _y ∈|y ₀ -R _y , y ₀ +R _y |

3. Move the sliding map t pixel by pixel within the sliding range, and calculate the similarity of the overlapping areas. The specific calculation process is as follows:

①Assuming that the moving position of the sliding map t in the search range is (x, y), calculate the maximum overlapping area range at this time, and obtain the maximum overlapping area g _f of the reference image and the overlapping area g _t of the sliding image. The calculation formula of the upper left corner coordinates (x _f , y _f ) of the overlapping area of the reference image is as follows:

The formula for calculating the coordinates (x′ _f , y′ _f ) of the lower right corner of the overlapping area of the reference image is as follows:

The calculation formula of the upper left corner coordinates (x _t , y _t ) of the overlapping area of the sliding image is as follows:

The calculation formula of the coordinates (x' _t , y' _t ) of the lower right corner of the overlapping area of the sliding image is as follows:

② Calculate the similarity between the corresponding overlapping regions g _f and g _t . Since the overlapping area has the characteristics of self-adaptive adjustment, the size of the included image is not fixed and will change with the position change of the sliding window. Therefore, a similarity measurement method independent of the image size needs to be used, and the normalized cross-correlation coefficient is used here. The similarity measurement is carried out, and the specific calculation formula is as follows:

Among them, (u, v) is the coordinate grid position of the g _f and g _t images, which is calculated pixel by pixel, and E(g _f ) and E(g _t ) are the gray mean values of the images g _f and g _t , respectively.

4. Record the position with the highest similarity of the maximum overlapping sliding window, which is the best matching position of the image.

The flow chart of the technical solution of the present invention is shown in FIG. 4 .

In addition, the above definitions of each method are not limited to the various specific ways mentioned in the embodiments, and those of ordinary skill in the art can simply modify or replace them, for example:

The normalized cross-correlation similarity measurement method adopted in step 3 may also adopt other similarity measurement methods that are independent of the number of pixels used, such as standard square deviation, standard correlation matching, and the like. As long as a similarity measurement method that is independent of the number of pixels is used to calculate the similarity of the maximum common area, it falls within the scope of the present invention.

The adaptive maximum sliding window matching method of the present invention will be described in detail below with reference to an example.

Select two images with a size of 510×800 to be matched, and take picture A as the benchmark, the exact matching position of picture B is (490, 5), and the size of the image overlapping area is 20×795, as shown in Figure 5. Since the width of the overlapping area is only 20 pixels, the general matching method cannot realize the matching processing of the two images.

Using the method of the present invention, the preset estimated matching position is (488, 7), and the search radius is set to 10 pixels. Through the adaptive maximum sliding window matching method, the similarity of each position is obtained as shown in Figure 6, and the optimal The matching position is (490, 5), which is consistent with the actual image offset position.

Based on the results obtained by matching, the results after stitching the images are shown in Figure 7, which meets the requirements of automatic matching of low-overlap images.

The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above-mentioned specific embodiments are only specific embodiments of the present invention, and are not intended to limit the present invention. Within the spirit and principle of the present invention, any modifications, equivalent replacements, improvements, etc. made should be included within the protection scope of the present invention.

Claims (9)

1. An adaptive maximum sliding window matching method is characterized by comprising the following steps:

dividing the image to be matched into a reference image and a slide image, and setting an estimated matching position (x) between the images to be matched₀，y₀) Setting the number of horizontal search pixels to R_xAnd the number of vertical search pixels is R_y；

Calculating the lateral sliding range T of the sliding image_xAnd a longitudinal sliding range T_y；

Moving the sliding image pixel by pixel in the sliding range, and calculating the similarity of the overlapping areas;

and recording the position with the highest similarity of the maximum overlapping sliding windows, namely the best matching position of the image.

2. Matching method according to claim 1, characterized in that the sliding image has a transverse sliding range T_xAnd a longitudinal sliding range T_yComprises the following steps:

3. the matching method according to claim 1, wherein the specific calculation process for calculating the similarity of the overlapping regions is as follows:

setting a slide image in a search rangeThe moving position is (x, y), the maximum overlapping area range at this time is calculated, and the maximum reference image overlapping area g is obtained_fOverlap area g with slide image_t；

Calculating the corresponding overlap area g_fAnd g_tThe similarity of (c).

4. Matching method according to claim 3, characterized in that said reference image overlap region upper left corner coordinate (x)_f，y_f) The calculation formula is as follows:

5. matching method according to claim 3, characterised in that said reference image overlap region lower right corner coordinate (x'_f，y′_f) The calculation formula is as follows:

6. matching method according to claim 3, characterized in that the sliding image overlap region has the coordinates (x) in the upper left corner_t，y_t) The calculation formula is as follows:

7. matching method according to claim 3, characterized in thatThe coordinates (x ') of the lower right corner of the overlapped area of the slide images'_t，y′_t) The calculation formula is as follows:

8. matching method according to claim 3, characterised in that said calculation of the corresponding overlapping area g_fAnd g_tThe similarity of (c) is measured using a normalized cross-correlation similarity measure, standard squared error, or standard correlation match.

9. The matching method according to claim 8, wherein the similarity measurement is performed by using the normalized cross-correlation coefficient, and the specific calculation formula is as follows:

wherein (u, v) is g_fAnd g_tThe coordinate grid position of the image is calculated by pixel-by-pixel, E (g)_f) And E (g)_t) Are respectively an image g_fAnd g_tThe gray level average of (1).

Images