Localized Multi-Channel Level Set Segmentation Combined With Texture Feature

Image segmentation, which is a key part in image processing, has important applications in medical image processing, remote sensing satellite image processing, and video processing.However, texture image segmentation is a difficult task in image processing. For complex background clutter images, getting an accurate object contour line is even more difficult.Classical region-based active contour model is Chan-Vese model. C-V model overcomes affects of the image segmentation coming from the image gradient information and has a certain resistance to noise. C-V model is a scalar field model and it uses only gray level information. To solve this problem, multi-channel CV model is extended. So it is suitable for vector image segmentation.Both models are global models. Taking into account the problems of uneven internal and external curve features in evolution and the image can not get local optimal solution. Thus, a multi-channel localized active contour model is proposed in the paper. It limits the evolution of the curve in the vicinity of a narrow band and ignores the curve evolution impact which caused by pixels away from the curve.we extract texture features by using three texture analysis methods.Meanwhile,we use GLCM, Gabor Filter and SCT to get texture information of different scales and direction. It combines gray level information of image as a model of multiple input channels.we do image segmentation experiments for images whose texture and background are complex. Finally an accurate object contour lines are got and validity and accuracy of the method is obtained.For target occlusion problem, it is also studied in the paper. we combines the algorithm and the priori shape information. Complete target contour can also be abtained by cutting up the incomplete target. Taking into account the image object translation, rotation and scaling issues, we adopt affine transformation. The model with prior shape information is validated through many experiments and it can solve the above problems.