An Edge Detection Method Based On Neural Networks

Edge is a basic character of images. Edge detection is base of image processing, and play an important part in image identify, cut, contract and so on. Now many kinds of edge detection method appear. Conventional arithmetic, for example Roberts, Sobel, can't be used by oneself. Many other subjects like Neural networks are used in image detection frequently. At the beginning edge detection is simple differential coefficient arithmetic, but now it becomes a mixed study including many subjects. Neural networks have abilities of self-study, adaptation and conmunication colligation. So they can simulate eyes to complete edge detections which simple edge arithmetics complete hardly. Using neural networks on edge detection is more and more popular. Based on existing theories, some work is done in this paper.(1) The theories of edge detection and neural network are studied. Characters of adult edge detection arithmetics are illustrated.(2) Date matrix two-dimensional bar code is used an imitating object. According the result of the experiment, the advantage and disadvantage are compared.(3) Pretreatment of the image is dong. After the bar code image is changed to a gray image from a color one, a few kinds of filters are discussed and simulated. Median filter is chose according simulating result. The circumgyratetion of the image is an important part. According the plumb projection, effective width of wave crest is got. Until the narrowest width is got, the image is circumrotated exactly. According the position of effective width, the plumb edge is defined. The level edge can be got in the same way.(4) The image is divorce into many parts. Each image part is inspected by disperse Hopfield neural networks. All inspecting results constitute the bar code information.(5) All work and result are summarized, and some disadvantages are present.