Values of the Red, Green and Blue are between 0 - the. Colors upgrade to darker towards to 0 and upgrade to lighter when towards to This situation is explained in the Cartesian coordinate system. The 0, 0, 0 origin point is black, and 1, 1, 1 point is white. Any color occurs as a result of the merger red, green, blue color with certain coefficients in the coordinate system. Gray color is above the white and black level, combining diagonal corners.
Thus, in the picture, there will remain only black, white and gray values. The process needs to be done before the image is converted into binary level.
The numerical value of the picture is reduced to two values with binary level. Thus, a 8 - bit image is converted into 2 - bit format. The threshold value must be determined for this conversion. Using a fixed threshold value is not correct because of external factors such as sunlight, shadows at real-plate images.
A distribution histogram is useful for calculating threshold value. If the pixel value in the image is greater then threshold value, then the pixel value is shown as "0"; and if the image pixel' value is less then threshold value, the pixel value is shown as "1". In this way the image is converted to the binary level. In each point, pixels on the image histogram and the number of pixels can be seen by using the image histogram. This image histogram does not show positons of the pixels on the image, but shows distribution of light-dark areas on the image.
The histogram distribution is required for the calculation of the threshold value. Histogram can be expressed mathematically with the following formula. In an image, if pixel values are clustered into a specific number, this clustering would be reduced by using histogram equalization. A histogram equalization is done for the plate image of Figure 3. A histogram equalization result in a license plate image is shown as figure 3. In this way, the object on the image will be separated from the background.
The data of histogram distribution is used for calculating the threshold value. Otsu method has been shown as a mathematical expression by figure 3. Figure Otsu Threshold Formula 3. This situation can be seen in Figure 3. Figure Threshold value of for the Image Results The iterative approach can be used instead of using this one.
According to the iterative approach, the algorithm is: 1. An initial treshold value is determined, t say Two different mean values are calculated for below m1 and above m2. The new treshold value is calculated. Elsewise, t would become tnew and it would also reiterate from step2. The image in the figure 3. Morphological operators can be applied to gray-level image and binary level image.
This matrix will be moved from left to right on the image. No processing will be done if all pixels of the matrix is at the background.
If any pixel of the matrix is on the object, the area under the matrix pixel and object intersection point is included to the background. If the objects in the image intersect the center of the matrix coordinate, the remaning space objects are combined at the bottom of the matrix.
In figure 3. All characters which are in plate image are generated by using the MSPaint program. Firstly, the recognition of these characters is checked. After this trial successfully fulfills the desired task, then the recognition of the real licence plate image is worked upon.
In MSPaint, the characters were generated as similar to real plate character sizes. Sizes of the characters which are generated are 36, 48, 78 punto. The success rate of character recognition of car license plates is lower than the one which is generated by using MSPaint. The reason for this is because the characters of license plate images do not have as sharp lines as the ones which are generated by using MSPaint.
Additionally, shadow and sunlight on the images might be the reason of this low recognition rate. Firstly pre-processing steps were done on the plate image for the recognition of the licence plate characters.
By doing this, the plate image noises were reduced as much as possible. The characters which are prepared by MSPaint do not need image processing steps. These stages are not required since characters are ideal in the image.
In this way color information in the image is removed. Plate image was converted to the gray level. In this way, the image plate is converted into the binary level.
Threshold value is difficult to be accepted since characters are created by using MSPaint. This value does not give successful results in the plate images. Otsu Thresholding method is used for plate characters. Although this method is successful in general-including the shadow-, the intense sunlight in the plate would not be successful. Therefore this method was abandoned. Then calculating the value of the dynamic threshold method is used.
Thus, noises such as shadow, sun light, "TR" logo were removed from the image plate. The image plate is converted to the binary system more clearly. In Figure 4. Figure 4. This operation is not required for recognition of characters prepared by using MSPaint. For this, the image plate which was converted to binary system is scanned in vertical and horizontal directions. Left and right borders of the each character are identified after vertical scanning, upper and lower borders of the each character are identified after horizontal scanning.
The upper and lower boundaries of license plate image boundaries may not be the same for each plate character. Upper and lower boundaries are found again for each plate character. The upper and lower boundaries are checked for each character after finding the character boundaries. Horizontal line is scanned again. The upper and lower boundaries of the characters may be changed after the scan result. Noises can be perceived as characters in the image plate.
The noise is deleted from the set of characters. Character recognition algorithm scans character line by line not only from left to right and from right to left, but also it scans column by column from top to bottom and bottom to top. The change in the character may be up, down, straight or jumping expressing sudden changes during scanning column by column characters from top to bottom, from bottom to top.
The size of the characters will be similar in the same plate. The size of characters must be matched up with the percentage slice to evaluate the percentage change in the character.
The width of the character is divided into hundred steps. On each step "1" is rated. While the width of the character is scanned step by step, the total value of these steps value will be "". Thus, change direction of the character has been found in the percentage. The same procedure is applied to the character's size. In table 4. Table 4. Thus, characters can be recognized.
Characters which change in sharp lines are not clear in the license plate image. There may be noise, changes in the shape of a ladder and recess ledges in the plate image. Therefore, problems have been experienced in character recognition. The changing edge of the image in figure 4. However, the lines might not be clear when the character which was generated from the license plate images is converted into binary system.
This situation is shown as an example in figure 4. In this case the threshold value is used for understanding the direction of change in character. For example, character "A" in figure 4. This illustration of the direction left, straight, left, straight of the character; left is proceeding as ordered in a way. These values are very different from the value of the character "A" which is stored in the database.
Hence, character recognition is very difficult. The character "A" which changes in figure 4. To resolve this situation difference value between the coordinates is used. Movement in the direction of the characters were not immediately decided if this value is smaller than the threshold value.
These unstable changes in the character are throwen into a building as a stack. The difference between the coordinates or characters till the end of the stack has continued to take action until it exceeds the threshold value. The difference between the first element coordinate of the stack and the last element coordinate of the stack is considered.
If the scanning process is from top to bottom and from bottom to top, then the difference between the line coordinates is to be obtained; and if the scanning process is from left to right and from right to left, then the difference between the column coordinates is to be obtained.
If difference value exceeds the threshold value, then the movement direction is determined depending on the difference sign. If the sign of the difference is positive the movement of direction is scanned from top to bottom and from bottom to top.
If Difference of the sign is negative then the movement direction is down. If the sign of the difference is positive, the movement of direction is right scanned from left to right and from right to left.
Movement direction is towars left if it is a negative sign By doing this, change of the edge of the "A" character in figure 4. In the same way the unstable situation for the character "L" in figure 4.
Another problem encountered in this step is about how the threshold value is calculated here. When the threshold value is selected as fixed, the results were not successful. Therefore, the threshold value is calculated dynamically depending on the length and width of the character. Thus, more successful results have been obtained. A different threshold value is used to for a sudden change in the direction of characters. For example, a sudden movement towards left is observed while scanning from right to left as in figure 4.
In order to see if the movement direction is towards left, the difference between the columns of the coordinates which are in between two points is obtained.
The direction of movement is determined when the difference is greater than the threshold value depending on the sign of the difference. Jumping points are important for their being distinguishing features for recognition of the character. Therefore, these points are given more points. However, the jumping points of the characters to be recognized and the jumping points of the same characters which are in the database may not be the same.
Hence, the score is increased depending on closeness of jump coordinates to the coordinates of the database. The further it gets from the jumping point, the less score it would receive. After a certain distance, a low score is given. The following are some examples for this code. The feature classes were created for each character for this data set. This feature classes were developed by using data-base classes Feature classes were created for the trial to recognize the characters.
Created feature classes are compared to previously created database class. By doing so, each the character's rate is calculated for trying in order to recognize a character. Similar rates are shown on the screen to the user. Each of the change rate is calculated for these changes and were evaluated over the score of a hundred.
This method is used in character recognition because this change will be different for each character in mind. NONE, null ; motionMap. TRAMP, null ; ch. UP, null ; motionMap. DOWN, null ; motionMap. LEFT, null ; motionMap. Figure 5. The plate image is selected from this screen. Processing results are shown in figure 5. The character which is desired to be displayed from the combo is selected. The results are listed on the screen as in figure 5. The most similar character set of the plate is displayed on the screen as in figure 5.
Developing software version 1. Developing application software IBM brand a desktop computer with Pentium 4 3. Test results are depicted in Table 6. It may be stated that character recognition algorithm ignores the curve in the middle while processing the character from the left side. What is more, algorithm assumes this curve as a noise and handle it as being straight.
Plates in Turkey should follow the rule that first and the second two characters must be numbers. The last two characters have to be elements of the numeric system. Accordingly, third character has to be one of the characters between A-Z.
This rule has also been implemented in the plate recognition program. As surface changes of the characters were considered in the algorithm, it was observed that noises within the character had no main effect on the overall identification. Figure 6. However, conversely, considering small changes as important did also cause misidentification.
In Table 6. Table 6. This algorithm can be improved in the future since the algorithm can easily be understood. If this algorithm is improved, then the correction rate will be higher. Neural network, support vector machine and other systems have complex mathematical algorithms.
Then improving and understanding is difficult for these algorithms. We can create three types character database. First is for small size charater, second is for normal size character and last one is for big size character. Then when the program detects the character, program can be shift to database automaticly which one is appropriate for detected character size.
Then we can improve the correction rate for this study. In this study, some problematic features like distance, light and corner are restricted. In future study, can be make solution for those problems. This study is interested only Turkish plate recognition. In future study can be interesting with international plate recognition. Finding plate location from digital image does not exist in this study. In future study it can be implemented also. CIT Proceedings of International Conference on Volume 7, Aug.
ISSPA ICET International Conference on Nov. ICTAI Digital Object Identifier Pacific-Asia Workshop on Volume 1, Dec. AMS ' CCDC GrC IITAW ' International Symposium on Dec. FGCN ' This dissertation, presents a morphology-based approach for the identification of a license plate in the image of a vehicle.
The recognition process deviates from the conventional approach of using Optical Character Recognition OCR systems and utilizes the concept of color coherence vectors [1]. Researchers have been proposed a variety of solutions for the problem of license plate identification and recognition in images. GOEL, P. The application of the Hough transform [3,4 and 5] has also been partially successful in reducing processing times for segmenting license plates in an image.
The utilization of enhanced edge-detection techniques [6 and 7] combined with others such as slope and projection evaluation is another interesting solution to this problem.
To attain faster processing speeds some systems decide on a threshold for the size of the license plate and the character regions within them. Then using fuzzy logic and neural network algorithms [8,9and 11] the character regions are segmented and the characters within them are identified. A slightly different approach to the segmentation problem is the mean shift segmentation method [10].
It identifies several candidate regions within a source image and utilizes features such as rectangularity, aspect ratio and edge density to determine whether the identified region is a license plate or not. All of the above research works strive to maintain a correct balance between the accuracy of the algorithm and its speed.
The morphology-based identification approach is highly accurate and the color coherence vector approach for recognition is extremely fast. A test application for the experimental evaluation of this proposed algorithm has been created using Microsoft Visual C. This algorithm and its experimental results shall be illustrated in detail in the sections ahead. The basic mathematical morphology [12] operations of dilation and erosion have been utilized. Firstly, an original image similar to the one shown in Figure 1 is converted to monochrome using two different thresholds.
The initial sample image The results of applying this operation on the source image are shown in Figure 2 and 3. Each of these images is utilized for the further steps of dilation and erosion. SINGH, et al Further, the image shown is Figure 3 is subjected to the dilation operation with a mask size of nine by nine and with white being the target color.
The white regions in the source image are expanded marginally as a result of this operation, the results of which are shown in Figure 4. This operation decreases the size of the white regions and helps in maintaining the larger white regions and removing the smaller ones.
The results of this operation are shown in Figure 5. The segmented region that best fits the above evaluation is then utilized for the next stage of algorithm.
This proposed identification algorithm works successfully even if the vehicle has been aligned at an angle of 30 degrees on either side, with the camera. A few sample results of the identification stage on images with varying orientation are shown in Figure 6. Figure 6. SINGH, et al 3. However, the processing time and accuracy of this technique are questionable.
This algorithm presented in this dissertation presents an extremely fast and accurate method of recognizing license plates. The algorithm can be termed as an illiterate one, in the sense that it does not extract the characters within the image but it recognizes the image as a whole. To build the initial database, images of the required license plates are preprocessed and their parameters are stored. During the recognition process these parameters are simply compared with those of the input image in constant time and the best match is retrieved.
Due to its static complexity it is an extremely fast technique for image recognition and the process of parameter extraction for the license plate image is based on the use of color coherence vectors [1]. This technique divides the entire grayscale image into buckets, where each buckets represents a range of gray color values.
This division of the image into buckets results in the segmentation of the characters from the rest of the unwanted background. Hence, during the comparison of these parameters the buckets representing the characters are compared to each other and the one which has the least overall error is displayed. Table 1 shows some of the successful and failed cases encountered during the testing of this proposed method. Sample results of the recognition algorithm 4.
The database has a sample size of sixty images and several test cases with variations in the alignment, illumination and size of the license plates were tested. Forty seven out of the sixty cases were successful and thirteen were not.
0コメント