Edge detection is the base of feature extraction

analysis and comprehension of image

the quality of which determines the results of subsequent processing. Thus it is an effort goal for people to find a kind of method that is insensitive to noise

precisely locates true edges and excludes false edges. This paper discusses and tests a novel algorithm of edge detection based on gray system theory. At first the characteristics of pixels are analyzed to decide non-edge referential sequence and the sequence to be compared

and then gray correlation degree of both sequences is employed to distinguish between non-edge pixels and edge pixels. Finally

the simulations prove that in binary image cases

the algorithm can obtain more precise edge pixels than some traditional edge detection algorithms

such as canny

log

sobel and zerocross

and tolerate some noise such as speckle

salt and gaussian. In grayscale image cases

it is unnecessary to change the referential sequence. Furthermore the quantity of edges can be controlled by adjusting the correlation degree threshold. So the technique is a new effective adjustable method of edge detection.