Target Segmentation Algorithm Based On SLIC And Superpixel

With the replacement of computer hardware and software,the image resources available to computers today are increasing geometrically.When massive image resources flood into our field of vision,in order to obtain the key information in the image,the computer must process these images accordingly so that the key information in the image is revealed.However,the lower accuracy and excessive execution time of traditional image processing methods obviously cannot cope with today's massive image resources.This undoubtedly brought new problems to the field of image processing,but also raised new challenges.Image segmentation bears the brunt as the foundation of computer vision.Traditional image segmentation algorithms have been unable to meet the growing demand in image processing.The shackles that the performance and efficiency of the image segmentation algorithm are incompatible are obviously the most stubborn bottlenecks in the field of image segmentation.In view of the high accuracy and real-time requirements for image segmentation in practical applications,this paper combines SLIC(Simple Linear Iterative Clustering Algorithm)and traditional area growth algorithms in the field of machine learning to analyze the two types of algorithms from an inclusive perspective The commonalities and differences between them,in order to find a balance between the "stable" of traditional algorithms and the "fast" of machine learning algorithms,so as to break the shackles of the existing image segmentation algorithm performance and efficiency.The research and discussion of SLIC(simple linear iterative clustering algorithm)and traditional area growth algorithms mainly include: transplanting the SLIC algorithm from the Lab color space to the grayscale image,and modifying the SLIC algorithm accordingly so that the SLIC algorithm can better fit Grayscale image;using the idea of superpixels in the SLIC algorithm to reduce the original scale of the image without losing image details,thereby reducing the difficulty of subsequent processing of image segmentation.Specifically,the research content and innovation results of this article mainly include the following two points:1.Application of optimized SLIC algorithm in grayscale imageThe SLIC algorithm itself is a superpixel segmentation algorithm tailored for the Lab color space,but because the gray image has certain similarity to the Lab color space in pixel attributes,this also provides a theoretical basis for the SLIC algorithm to be transplanted tothe gray image.The traditional SLIC algorithm is too cumbersome in the selection of the initial clustering center.This series of tedious operations is particularly important for the Lab color space,but transplanting the SLIC algorithm to the grayscale image is a little bit cumbersome for the initial clustering center.Was superfluous.In view of the characteristics of the gray-scale image itself,this paper attempts to introduce the variance between pixels and use the variance between pixels to partially replace the gradient descent operation of the clustering center in the original SLIC algorithm.Such changes to the SLIC algorithm greatly increase the speed of the SLIC algorithm on grayscale images without affecting the accuracy of superpixel segmentation.2.Introducing superpixels into traditional area growth algorithmsThe segmentation result of the traditional region growth algorithm depends on the selection of seed points,and the image itself noise and uneven gray value are easy to form a segmentation hole in the segmentation target process.An improved region growth algorithm based on superpixels is proposed for the above problems.Use Laplace sharpening to enhance the boundary of the target to be segmented,and then use SLIC(Simple Linear Iterative Clustering Algorithm)super-pixel segmentation to divide the original image into several irregular regions based on the characteristics of similar pixel grayscale,and establish an irregular region In the undirected weighted graph,select the seed area,and grow the area in the unit of divided irregular regions according to the undirected weighted graph.Finally,the area is grown in pixels at the edge of the segmentation target,and the boundary is refined.Compared with the traditional region growth algorithm,the improved algorithm is less affected by the selection of seed points in the segmentation results,and can effectively solve the problem of segmentation holes.Compared with typical algorithms such as clustering segmentation and otsu(maximum between-class variance)threshold segmentation method,this algorithm has obvious advantages in segmentation accuracy.