An Improved method for Harris corner detection is proposed while considering that there are two limitations exits in the Harris method. The first is its low localization to two kinds of corners(T and diagonal T). The second is its low efficiency. The improved method compare the value between the kernel pixel and pixels in the kernel pixel’s 8-neigbourghood while in which the value demonstrate the number of pixels that are similar to the kernel pixel. As a result of comparison
some of the pixels are reserved for the next step computation. If the value of the Harris corner response function(CRF) is bigger than the threshold value
the pixel will be considered as corner finally. The experiment show that the improved method achieves a better performance.