Adaptive Threshold Segmentation Technology And Its Application In Industrial Visual Inspection

Image threshold is an important pre-processing step in the field of industrial visualinspection, its main purpose is to separate target areas of interest from complex backgroundfor scene analysis and target recognition. Currently, threshold technology is still facing lowaccuracy, poor uneven illumination resistance, not universal problems. To solve the aboveproblems, global adaptive algorithm and local adaptive algorithm are studied, and analyzesthe advantages and limitations of adaptive algorithm. The traditional global algorithm ispresented for further improved scheme, and local threshold based on grayscale wave foruneven illumination image is proposed. In this paper, the main research results are as follows:(1) For the existing gray level-average gray level histogram exists approximatelymisclassification, a global adaptive algorithm based on gray-gradient2D symmetric Tsalliscross entropy is proposed. Firstly, neighborhood gradient information is used as the seconddimensional feature to build a gray-gradient histogram, more fully consider the internal pointsof background and target class. Simultaneously, criterion function of symmetric Tsallis crossentropy threshold based on the histogram zoning are derived. Fast recursive algorithm isintroduced to reduce computational complexity of correlation formula for criterion function.Finally, criterion function as a fitness function, and chaotic niche particle swarm algorithmbased on tent map is used to search for the optimal threshold vector, and further improve algo-rithm accuracy and real-time. The experimental results show that this method can make imageedges more accurate, more uniform gray within the class, and real-time is greatly improved.(2) Taking into account industrial images are often disturbed uneven illumination,compared to global algorithm, local algorithm has better results, but local algorithm existsproblems such as poor robustness, real-time slowly. To solve the above problems, accordingto Water Flow model theory, starting from image3D gray surface, local threshold algorithmbased on grayscale wave for non-uniform illumination image is proposed. Firstly, it extractsimage grayscale wave curve in the horizontal and vertical direction, and iteratively search forlarger scale peaks point and troughs point of each curve to meet a given wave amplitudethreshold. Then, each pair of alternating peaks or troughs is used to calculate floatingthreshold to designate the target and background pixels attribution. Finally, it takes theintersection of threshold image in two directions to obtain the final segmented image.Simulation experiments show that this method reduce the impact of non-uniform illuminationon the image segmentation in a large extent, and improve the effect of image segmentation.(3) In order to fully verify the actual performance of algorithm, collecting a lot of imagesto build your own image library, and testing and analysis the above algorithms. Actual testsshow that improved2D symmetric Tsallis cross entropy algorithm can be more highlight theedge details of target in more complex situations. Local threshold algorithm based on gray-scale wave can handle industrial uneven illumination image, and by adjusting the amplitudethreshold to control sensitive degree of wave, in order to adapt to the different environment.