Image segmentation algorithm based on mean shift with high boundary precision and anti-noise performance

There is not an uniform standard of determining the bandwidth of window of kernel function for mean shift.Images are divided into three sub-images according to the orientation information of edges detected by canny operator.Regular edge sub-image contained regular edge pixels.Nonregular edge sub-image is composed of edge pixels whose orientation changed sharply.Non-edge sub-image consisted of inner pixels in smooth region and noise pixels.The bandwidth of kernel function window in regular edge sub-image and non-edge sub-image is large in order to smooth inner region and improve anti-noise performance.The bandwidth of kernel function window in nonregular edge sub-image is small so that the segmented image had more accurate boundary.The experimental results show that the method of selection bandwidth of kernel function window in different sub-image is effective to metal fracture images for more accurate boundary and better anti-noise performance.