Study On Industrial X-Ray Image Sharpening Technology Algorithm

In recent years, along with the rapid development of technology, image processing isused in a rapidly increasing host of human life, especially for applications inindustrial inspection, national defense and military, biomedicine, security check andother fields. In industrial production, the industrial X-ray detection technology draw moreand more attention, because of that the defect detection and recognition for products internalstructure is very necessary for product quality ensurance. However, in the industrial X-raydetection system, the influence of detection equipment, imaging environment, the complexityof the workpiece to be checked and other factors often lead to the quality degradation ofcollected images, such as low image resolution, blurred edges or details, low image contrastand so on. All of these factors would directly affect the judgment on the defect detection andrecognition for products internal structure. Therefore, in order to get more useful informationfrom the interested image and improve the accuracy of product testing, it is need to do someimage enhancement processing, so that images with higher readability and clarity could beobtained.In this paper, some domestic and foreign existing image enhancement algorithms andother basic methods were expound. Moreover, on the premise of the the analysis for theirexistence problems and defects in the practical application, some new methods and new ideaswhich based on the detail enhancement technology were proposed, and achieved good results.The main contents of this paper are as follows.1. Based on an in-depth understanding of mathematical morphology and relatedproperties, an adaptive image enhancement algorithm with variable weighted matching basedon morphological (VWMM) was studied in this work. In order to improve extractionaccuracy for details and enhance the sensitivity of details of different directions in imageenhancement process, VWMM took the effect of structural elements on image processinginto consideration, and constructed comprehensive multi-scale structural elements. VWMM brokes the idea of that the detail weighted in each direction is taken average in traditionalmorphology method. When it comes to the detail processing, the detail weights of differentdirections were adjusted by the local gray properties, and further more, the gain functioncould be adjusted adaptively through the extracted details. The proposed VWMM algorithmgot a more comprehensive view of the image autocorrelation and improved the pertinenceand flexibility in image processing.2. A new image detail enhancement sharpening algorithm was proposed to solve theproblem that the existing improved algorithms only consider the intensity information alongwith change of the grayscale value. From the perspective of the intensity and frequency ofgray scale change, the new method combined the local gradient and local complexity andconstructed an adaptive gain function, which replaced the thoughts that traditional algorithmsdefined weights rely only on the intensity change of image grayscale. On the basis of thisthought, the image ENI(Edge pixels, Noisy pixels and Interior pixels) was uitlized in theour further improved algorithm, which could distinguish the the edge pixels, the noise pixelsand the internal pixels more precisely and further refined the detail processing. For the grayupheaval region and rich level region, a variety of local statistical parameters were utilized toreplace the single local statistics adjustment gain function, which could offer more efficientinformation for the edge-and detail-enhancement.3. An unsharp masking enhancement algorithm based on double smoothing filter andgeneralized linear operation will be studied after detailed analysis of the main characteristicsand defects of unsharp masking algorithm is provided. Due to generalized system can avoidthis situation that superposition process will lead to data overflow, this proposed algorithm isa new framework for image enhancement that is designed in the generalized linear system.This method analyzed the difference between Gaussian filter and edge preserving filter andutilized the advantages ofL0norm gradient minimization smoothing filter in smoothingtexture and preserving edges, which weaken halo phenomenon and gradient reversalphenomenon produced in the process of enhancing. Firstly, the high frequency part without noises in flat area will be obtained after the difference between these two results that theoriginal image is respectively filtered byL0norm gradient minimization filter and Gaussiansmoothing filter. Then, the information of this high frequency is used to get enhanced resultaccording to the un-sharp masking formula, which will decline human eye’s sensitivity to thevisibility of noise in flat area and obtain enhanced image with better visual effect.4. As the patch similarity can keep the information effectively, we studied a sharpeningmask algorithm of superposition of anisotropic structure weight based on patch similarity.The algorithm making full use of the local characteristics of the image patch, describes andextracts details of the image from a new point of view. As the patch similarity has strongability of expressing the local information, and contains more information, so applying thepatch similarity to the detail enhancement, can highlight the weak details of the image. At thesame time, the patch similarity can avoid the sensitivity of the isolated points or lines, andreduce the effect of noise.