Level Set Image Segmentation Based On Local And Global Information

Although medical images and radar images are grayscale images,due to the detection method of the instrument,the resulting images have blurred edges and complex energy distribution.Natural color images often have the problem of multiple targets in the same color and different colors in the same target,which increases the difficulty of recognition.The level set algorithm can be used as a preprocessing step or as the final segmentation result.It can also add preprocessing steps such as clustering and threshold to the level set model,and can also be integrated into other segmentation algorithms(such as k means).The level set algorithm has good topology recognition ability and stable closed contour retention ability.In order to understand the characteristics of the level set algorithm in more detail and improve the performance of the algorithm,this article mainly does the following work:(1)While summarizing the level set algorithm in the past decade,this paper also makes an experimental comparison of the performance and advantages and disadvantages of common algorithms.The experimental images are medical images and images with uneven intensity.From the performance point of view,the level set algorithm has greatly improved the segmentation speed and stability,and has a significant effect in the anti-noise optimization.After the comparison,the possible directions of the future level set development are summarized.(2)The Taylor expansion numerical optimization and offset domain correction are substituted into the traditional level set energy term to improve the stability of segmentation,and a new energy term based on the image numerical mean square error mapping function is added to the level set function.By improving the existing algorithms,a level set algorithm based on global energy optimization and local energy segmentation is proposed,which optimizes the algorithm from various aspects such as color space images,function polyphase processing,and parameter initialization.The experimental results of the proposed algorithm are analyzed and compared in Chapter 5.Due to the multi-dimensional processing of functions in segmentation,the segmentation objects are divided into grayscale images and color images.You can see the output from the results.In some images,the results of the proposed algorithm are better than the compared algorithms.It will also propose algorithms to try to apply to images in different scenes.(3)The proposed algorithm is applied to gray-scale images and color images respectively,and the results are compared with other algorithms in terms of iteration times,segmentation time,Jaccard coefficients,etc.,showing the superiority of the proposed algorithm.