Research On Target Enhancement And Detection Based On X-Ray Image

In contemporary society, X-ray real-time digital imaging detection technology has been developed rapidly with the improvement of imaging technology of X-ray flat panel detector. Target detection and recognition based on image has become an important direction of X-ray nondestructive detection. Automatic defect detection, recognition and evaluation based on digital image has been widely used. However, there are still some problems in the existing detection methods. Firstly, traditional detection will lead to underexposure within the thick part while overexposure within the thin part under a single voltage, in which case we are unable to get the whole information of the subject. Secondly, when applying the traditional X-ray examination to the field of wheel detection, the final result counts largely on artificial flaw on the X-ray image. This wastes a huge amount of effort and the accuracy is fairly low. So this paper is dedicated to solving the above two problems.For the first problem, this paper references the multi-exposure technology which usually used in visible image field. Firstly the image of work piece using multi-voltage imaging can be got. Then this paper constructed the fusion weighted image using the grayscale, local contrast and local information entropy. In the process of fusion, this paper first conducted multi-scale decomposition on every voltage image and then fused them altogether to get an accurate X-ray image with high dynamic rage (HDR), which contains the complete information of the work piece. At last, in order to display the HDR image in traditional screens, the color mapping method was improved, which intensified the local details and finally got an image with high contrast and clear structure.For the second problem, this paper firstly proposed an image-based automated labeling method of region of interest (ROI). This method is more accurate compared with traditional ROI labeling, thus improved the correctness of flaw detection. Secondly, this paper introduced integrated image into traditional flaw extracting algorithm. This introduction largely improves the efficiency of algorithm because the integration markedly decreases the time of calculating the local average of image, which occurs frequently in the whole calculations. Thirdly, by using the connected-component labeling algorithm to eliminate the pseudo-flaws caused by pulse noise, this paper cut down the misjudgments. Last but not least, this paper modified the shape around the flaw region with the help of topological calculation in order to get a more authentic shape of the flaw region, thus to provide a result reliable enough to conduct the following procedures such as flaw classify.