Slict: Computing Texture-sensitive Superpixels In Natural Images And Medical Images Respectively

The purpose of image segmentation is dividing an image into several regions which have same or similar color and texture.Replacing pixels by superpixels as the input of subsequent image processing can improve processing speed remarkably.Besides,a superpixel algorithm that is more precise can also improve the precision of subsequent image processing more observably.Among those typical superpixel algorithm,SLIC catches attention owing to its simple algorithm,higher speed and good performance.But SLIC does not perform well in some regions with similar color,because SLIC are not sensitive to texture and lack of orientation when merging orphan superpixels.To overcome this shortage,we propose a new superpixel algorithm called Simple Linear Iterative Clustering based on Texture(SLICT).SLICT is based on a kind of texture feature without texture deviation,that texture feature which can be obtained from the original image by improved LBP algorithm proposed in this paper.Then we combine this texture feature with traditional SLIC to calculate superpixels.The experiment result shows that the texture feature obtained by the improved LBP algorithm contains less noise and shows more continuity and discontinuity in similar and dissimilar regions respectively.Furthermore,SLICT are more sensitive to boundary and texture as well as the precision are higher.In this paper,we introduce the traditional SLIC and LBP algorithm,analyze their advantages and drawbacks,and propose their improving methods respectively.The main work in this paper is as follows:(1)In order to solve the problem called texture deviation in traditional LBP algorithm,we propose that it can be overcame by adding valid threshold in the equation calculating LBP value.Experiment result shows that the improved LBP algorithm are more robust to noise and can handle with texture deviation better.(2)To verify the effectiveness of the improved LBP algorithm on extracting texture feature in medical and natural images,we firstly discuss the texture performance in these two kinds of images.Secondly,we propose that using extremum-suppression method can get better texture feature in medical images and using non-extremum-suppression method can get better texture feature in natural images.The experiment result shows that the improved LBP algorithm with or without extremum-suppression can get better texture feature in medical and natural images respectively.(3)To overcome the insensitiveness to texture of SLIC algorithm,we propose a method by adding texture distance in iterative clustering.The experiment result shows that using the updated distance equation in clustering can generate a better result which is moresensitive to subtle boundary and subtle texture.(4)To overcome the chaos in merging superpixels of SLIC algorithm,we propose a method that using texture feature as the merging condition.Then we discuss the difference between using 4-neighbor and 8-neighbor searching strategy in merging process.The experiment result shows that using 8-neighbor searching strategy can generate more texture-sensitive result which can adhere to boundary more tightly.