Study On Enhancement And Workpiece Characters Recognition Algorithm For Casting DR Images

Non-destructive testing system with X-ray Digital Radiography (DR) for castings can effectively detect the internal defects. There are often specific serial numbers on the casting, which are known as the workpiece characters. And these characters are also reflected in the casting DR images. Recognizing the workpiece characters of casting DR image has great significance for imputing the detection information automatically and reducing the labor intensity. There are many similarities between casting workpiece characters recognition and license plate character recognition. After comparing the license plate character recognition system with the casting workpiece characters recognition system, this paper designed a solution of the casting workpiece characters recognition based on the inherent characteristics of the casting DR images. We focus on the image enhancement and binarization algorithms of the character part, workpiece character segmentation algorithm, and preliminary study the workpiece character recognition algorithm.Due to the influence of casting's thickness and other factors, there are some problems in the character region of the casting DR image, such as not obvious to distinguish characters from background. And this situation affects the later recognition and analysis. Therefore, in order to identify the characters precisely, a further image processing is necessary, such as image enhancement. For casting DR images, we could make use of the enhancement method of non-uniform illumination image. After investigating the commonly used enhancement methods, we use the additive model to eliminate the impact of dark region. Use image block and interpolation to obtain the approximate light background image of the character region of the casting DR image. Since the image subtraction between original image and background image makes the gray range of the result image to become narrow. This makes the result image become very dark. So we use gamma correction to adjust the gray range. It can enhance the contrast of the DR image, and the result is good for binary.After researching some commonly used binary methods, we found that the local threshold algorithm is suitable for the enhanced character images, however these methods such as Bernsen method and Nilblack method are not so effective in DR images as expected. The second edge extraction algorithm could not only wipe of the influence of the background, but also could complete the binarization. So we make use of this method to the enhancement character image. But there are fracture strokes in the result image of this method. Thus we improve it. The improved method could reduce the number of fracture strokes and get a good binary image.Based on the characteristics of the binary images, character segmentation is divided into horizontal segmentation and vertical segmentation. We use horizontal projection of the right half part of binary images to realize horizontal segmentation. And then, as vertical segmentation, for these images which have obvious alternations between characters, we directly make use of vertical projection to realize vertical segmentation; but for those images which have obvious interference, we would like to use other method. Considering the wavelet transform can analyze the signal on different scales, and the scales can be chosen for different purpose. So wavelet transform is used to extract the detail information of the image's column sum, and then determining the column space between two characters. As a result each character could be segmented. Grid method is adopted to extract features of number character, and use template matching to achieve recognition. For actual casting DR image, experiments show that these algorithms can obtain good results.