A Novel Image Segmentation Based on Pixel Probability Measurement Model

Natural image segmentation is an important Held in image and computer vision processing. Custer-based image segmentation algorithm, which is the main method of the unsupervised image segmentation algorithm is the most widely used segmentation method. However, these methods generally use pixel-based feature extraction, which directly leads to a poor fit between the segmentation result and the boundary. To mitigate this problem, the paper proposes a serialized processing process for image segmentation. It uses multiple clues such as the brightness and texture of image pixels. A sequence of image pixel regions with merge processing steps has been designed. They gradually generate segment-able regions. In each step of the process, probability metrics have been used to quantitatively analyze the similarity of feature clues in adjacent regions. This determines whether they belong to the same segmented region or not. The image brightness and texture distribution of each area and the surrounding local area is the main clue combination of the probability formula that has been designed. The geometric shape of the area is another important clue of a priori design. At the end of the algorithm, the combined posterior of the mixed expert formula has been designed through feature clues such as brightness and texture. Combining this probability measurement method, a complete image-level segmentation algorithm is formed. Its complexity is only affected by the number of image pixels, and the relationship is linear. Almost no intervention is needed to adjust the parameter values during the execution of the algorithm. At the end of the article, the experimental results on the custom test data set (collected from the standard data set) show that the algorithm solves the above existing problems and has achieved good experimental results.