CN101894255B - A container number location method based on wavelet transform - Google Patents

Abstract

A container number positioning method based on wavelet transformation belongs to the technical field of image processing. The invention utilizes the gray information of the original picture of the container number to be identified, carries out median filtering, binaryzation and morphological processing on high-frequency information after one-dimensional wavelet transformation, extracts a connected region by adopting a 4-neighborhood connected method, judges the region of the container number in the original picture according to the length and the width of the connected region and the position relation among the connected regions, and finally detects the inclination angle of the number by utilizing Hough transformation and carries out rotation correction. The invention adjusts the high-frequency coefficient binarization threshold value so as to realize the positioning of all container number candidate areas, thereby avoiding the missing positioning of the container numbers and improving the accurate positioning rate of the container numbers. The invention is not sensitive to noise, and small characters possibly appearing on the container have no influence on the positioning result of the container number.

Description

A container number location method based on wavelet transform

technical field

The invention belongs to the field of image processing, and mainly relates to a number positioning technology in a container number recognition system.

Background technique

With the rapid development of the economy, the throughput of major ports is increasing. Efficient management of containers on ports has become an urgent problem to be solved, which is related to the efficiency of containers leaving and entering the port, and also directly related to the throughput that the port can withstand. However, the port now uses manual methods to register the incoming and outgoing containers and go through relevant procedures, which seriously affects the efficiency of container incoming and outgoing, and also affects the capacity of the port. Moreover, the port needs to work around the clock, so it is difficult to ensure that the port staff will not make mistakes due to factors such as fatigue and mood. In order to efficiently and accurately manage the entry and exit of port containers, the research on the automatic identification system of containers becomes more urgent. With the container number automatic identification system, people can automatically register and go through relevant procedures for containers entering and leaving the port around the clock, so as to shorten the detention time of containers in the port, improve the efficiency of container entry and exit, and ensure the registration of container numbers Work is not affected by people's mood, fatigue and other factors.

A complete container identification system requires at least three steps: 1) container number positioning, 2) container number segmentation, and 3) container number identification. Due to the various arrangements of the container numbers and the various color backgrounds of the containers, it is more difficult to locate the container numbers.

At present, the existing container number positioning methods mainly include:

1. First find suspicious character candidate boxes in the picture, and then perform screening to remove character candidate blocks that are not container numbers as much as possible. At this time, further use the center point of the character candidate block as the input of hough transform to locate A straight line passing through as many character candidate blocks as possible to locate the container characters.

2. First, scan the row or column data of the picture, and mark the number of pixels whose gradient difference is greater than a certain threshold in a row or column. If the number of such pixels in a row is greater than a certain threshold, the row Suspicious line as container number. If no suspicious line is found in the entire image, lower the threshold and rescan to find it. After finding the suspicious line of the container number, further confirm the suspicious line, that is, scan and mark the consecutive lines of data below the suspicious line, and confirm if the marked point of each line is similar to the marked point of the previous line The line where the container number of the suspicious behavior is located, which completes the rough positioning of the container number. Then the system generates arrangement rules based on the existing arrangement of container numbers, and then compares the rules with the picture data after rough positioning, so as to locate the correct container number plate.

The above algorithms all have problems to a certain extent in container number positioning. The algorithm for locating the suspicious character candidate frame in the picture is not described in the positioning algorithm 1, and the positioning algorithm for the character candidate frame cannot be known. However, it is very difficult to accurately locate each suspicious character candidate box from the original picture, and the accuracy rate will also vary due to the quality of the picture, which results in a low accuracy rate of the final container number. Positioning Algorithm 2 uses the gray level difference in the picture to roughly position the container number plate, which determines that the method is more sensitive to noise.

Contents of the invention

The invention provides a container number location method based on wavelet transform. The method can locate the container number in a complex background, and is not sensitive to noise, and the small characters that may appear on the container have no influence on the container number location result.

Detailed technical scheme of the present invention is as follows:

A container number location method based on wavelet transform, as shown in Figure 1, comprises the following steps:

Step 1: Collect the original image of the container number to be identified and convert it into a grayscale image.

Since there is no fixed color match between the background color of the container and the container number, the container number cannot be located based on the color value, so it is necessary to convert the original image of the container number to be identified into a grayscale image. If the collected original picture of the container number to be identified is in RGB format, the conversion formula for converting it into a grayscale format picture is: gray=0.229×R+0.587×G+0.114×B, where gray represents a certain value in the grayscale format picture. R, G, and B represent the pixel values of the red, green, and blue channels of the pixel in the RGB format picture, respectively.

Step 2: Use Harr wavelet to perform one-dimensional wavelet transform on the gray scale image obtained in step 1.

Step 3: The high-frequency coefficients obtained after wavelet transformation are combined into a picture, and median filtering is performed on the picture.

Median filtering is to remove the influence of noise, and it has a good effect on removing isolated noise in pictures. The window of the median filter is 3×3 pixels or 5×5 pixels in size.

Step 4: Binarize the image after median filtering.

Since 90% of the high-frequency coefficients after median filtering are concentrated below 30, the binarization threshold during binarization processing can be determined between 3-5.

Step 5: Perform morphological processing on the binarized image.

Using a structure that is half the size of the character of the container number as a morphologically processed structure first corrodes the binary image and then expands it.

Due to the drastic transformation of the gray scale of the container number characters, the wavelet transform has a large high-frequency coefficient. After binarization, the container number is binarized to 1, while other background places are binarized to 0, but there may be one or several characters that are binarized with other characters If there is no connection, some noise areas will also appear. Although this type of noise area is binarized to 1, the area is small and isolated far away. Using the morphological method to first corrode the picture and then dilate it, the noisy area can be removed and the broken container numbers can be connected at the same time.

Step 6: Extract connected regions in the binarized image after morphological processing.

The 4-neighborhood connectivity method is used to mark the connected regions in the binarized image after morphological processing, and the size and position information of each connected region is counted one by one.

Step 7: Judging the connected areas extracted in step 6, and finding the area corresponding to the container number in the original picture.

The container number arrangement includes one column, two columns, one line and two rows. Due to the diversity of the arrangement of the container number itself, it is necessary to analyze and judge each connected area to determine which connected area is the real number area. In actual operation, judge the area of the container number in the original picture according to the length, width and positional relationship between the connected areas: first select the longest connected area, and judge whether the area reaches the container number in a row or column number arrangement Length, if it is, the area corresponding to the longest connected area in the original picture is the container number area; if not, the area corresponding to the longest connected area plus the second longest connected area in the original picture is the container number area.

Step 8: Detect the inclination of the container number area obtained in step 7, and make corresponding rotation adjustments.

First extract the edge points of the container number area obtained in step 7, and then use the Hough transform to detect the inclination of the container number area. If the inclination exceeds 2°, use the interpolation algorithm to rotate it to the horizontal.

Through the above steps, the area where the number is located can be located in the original picture of the container number to be identified. Of course, after the above steps, the container number is only roughly located. To achieve the identification of the container number, it is necessary to accurately locate the located container number plate and separate each character. The present invention uses wavelet transform to find the character area of the container. Since the grayscale transformation of the character area is relatively large, the high-frequency coefficient after the transformation is relatively large, so the present invention can locate the area where the container number is located by using the high-frequency coefficient. The morphological processing can also solve the problem of image noise very well. The invention can adjust the binarization threshold of high-frequency coefficients so as to locate all candidate areas of container numbers, thereby avoiding missed positioning of container numbers and improving the accurate positioning rate of container numbers.

Description of drawings

Fig. 1 is a schematic flow chart of the present invention.

Detailed ways

A container number location method based on wavelet transform, as shown in Figure 1, comprises the following steps:

Step 1: Collect the original image of the container number to be identified and convert it into a grayscale image.

Since there is no fixed color match between the background color of the container and the container number, the container number cannot be located based on the color value, so it is necessary to convert the original image of the container number to be identified into a grayscale image. If the collected original picture of the container number to be identified is in RGB format, the conversion formula for converting it into a grayscale format picture is: gray=0.229×R+0.587×G+0.114×B, where gray represents a certain value in the grayscale format picture. R, G, and B represent the pixel values of the red, green, and blue channels of the pixel in the RGB format picture, respectively.

Step 2: Use Harr wavelet to perform one-dimensional wavelet transform on the gray scale image obtained in step 1.

Step 3: The high-frequency coefficients obtained after wavelet transformation are combined into a picture, and median filtering is performed on the picture.

Median filtering is to remove the influence of noise, and it has a good effect on removing isolated noise in pictures. The window of the median filter is 3×3 pixels or 5×5 pixels in size.

Step 4: Binarize the image after median filtering.

Since 90% of the high-frequency coefficients after median filtering are concentrated below 30, the binarization threshold during binarization processing can be determined between 3-5.

Step 5: Perform morphological processing on the binarized image.

Using a structure that is half the size of the character of the container number as a morphologically processed structure first corrodes the binary image and then expands it.

Due to the drastic transformation of the gray scale of the container number characters, the wavelet transform has a large high-frequency coefficient. After binarization, the container number is binarized to 1, while other background places are binarized to 0, but there may be one or several characters that are binarized with other characters If there is no connection, some noise areas will also appear. Although this type of noise area is binarized to 1, the area is small and isolated far away. Using the morphological method to first corrode the picture and then dilate it, the noisy area can be removed and the broken container numbers can be connected at the same time.

Step 6: Extract connected regions in the binarized image after morphological processing.

The 4-neighborhood connectivity method is used to mark the connected regions in the binarized image after morphological processing, and the size and position information of each connected region is counted one by one.

Step 7: Judging the connected areas extracted in step 6, and finding the area corresponding to the container number in the original picture.

The container number arrangement includes one column, two columns, one line and two rows. Due to the diversity of the arrangement of the container number itself, it is necessary to analyze and judge each connected area to determine which connected area is the real number area. In actual operation, judge the area of the container number in the original picture according to the length, width and positional relationship between the connected areas: first select the longest connected area, and judge whether the area reaches the container number in a row or column number arrangement Length, if it is, the area corresponding to the longest connected area in the original picture is the container number area; if not, the area corresponding to the longest connected area plus the second longest connected area in the original picture is the container number area.

Step 8: Detect the inclination of the container number area obtained in step 7, and make corresponding rotation adjustments.

First extract the edge points of the container number area obtained in step 7, and then use the Hough transform to detect the inclination of the container number area. If the inclination exceeds 2°, use the interpolation algorithm to rotate it to the horizontal.

Through the above steps, the area where the number is located can be located in the original picture of the container number to be identified.

Claims (2)

1. container number positioning method based on wavelet transformation may further comprise the steps:

Step 1: the former picture of container number to be identified of collection, and convert thereof into the gray scale format picture; If the former picture of gathering of container number to be identified is the rgb format picture; The conversion formula that converts thereof into the gray scale format picture is: gray=0.229 * R+0.587 * G+0.114 * B; Wherein gray representes certain gray values of pixel points in the gray scale format picture, and R, G and B represent the pixel value of redness, green and blue three passages of this pixel in the rgb format picture respectively;

Step 2: select the harr small echo for use, the gray scale format picture of step 1 gained is carried out one-dimensional wavelet transform;

Step 3: the high frequency coefficient that obtains behind the wavelet transformation is formed picture, this picture is carried out medium filtering;

Step 4: the picture behind the medium filtering is carried out binary conversion treatment; Binary-state threshold during binary conversion treatment is confirmed between 3-5;

Step 5: the binaryzation picture is carried out morphology handle;

The structure that employing is handled as morphology with respect to half big or small structure of container number code character corrodes the advanced row of binaryzation picture and then expands;

Step 6: extract morphology and handle the connected region in the binaryzation picture of back;

Adopt 4-neighborhood communicating method to mark morphology and handle the connected region in the binaryzation picture of back, and count size, the positional information of each connected region one by one;

Step 7: the connected region to step 6 extracts is judged, and is found container number The corresponding area in former picture;

Position according between length, width and the connected region of connected region concerns the zone of judging container number in the former picture: at first choose the longest connected region; Judge whether this zone reaches the container number code length of a delegation or a column number arrangement mode, if then this longest connected region The corresponding area in former picture is exactly the container number zone; If not, then the longest connected region adds that second long connected region The corresponding area in former picture is exactly the container number zone;

Step 8: detect the degree of tilt in step 7 gained container number zone, and do corresponding rotation adjustment;

At first extract step 7 gained container number edges of regions point, utilize Hough transformation to detect the degree of tilt in container number zone then,, then utilize interpolation algorithm that it is rotated to level if degree of tilt surpasses 2 °.

2. the container number positioning method based on wavelet transformation according to claim 1 is characterized in that, the window of step 3 medium filtering is 3 * 3 pixels or 5 * 5 pixel sizes.

Images