An improved Sobel algorithm based on median filter

This paper provides an advanced algorithm based on classical Sobel operator whose defect is rough effect for edge detection and extremely sensitive to the noise. This paper proposes an improved algorithm which used the improved Sobel operator with a combination of median filtering method. This algorithm makes the edge detection for the Salt and pepper noise images and it effectively overcome the problem that the Sobel operator is only sensitive to vertical and horizontal direction. It combines the advantage of the median filter to remove salt-pepper noise. The effect of the experiment is better than the classical edge detection method.

I. INTRODUCTION

The edge is the essential feature of the image, and it is the collection of pixels that the pixel grayscales have step changes or roof-like changes around the image. It exists between background and objectives, objectives and objectives, regional and regional, primitives and primitives. The edges can sketch out the target object, and contains a wealth of information. It is an important property of extracting image characteristics in image segmentation, identification and analysis. Edge has the direction and magnitude two characteristics. Along the edge direction, pixel value changes more gently, while perpendicular to the edge direction, pixel value changes more dramatic, may shows step-like, may also shows vertical roof. So the edges can be divided into two kinds: one is step-like edge, the pixel gray on its both sides is significantly different. The other is roof-like edge. It is located in the turning point where the gray value changes from increase to decrease. For the steplike edge, second order directional present zero-crossing at the edge; while for the roof-like edge, second order directional takes the extreme at the edge. The image edge detection provides the fundamental basis for to determining the edge point by the extreme of first derivative or second derivative zero-crossing information of image, and is an important image segmentation method. If a pixel falls on an object boundary, its neighborhood will become a band whose gray-scale changes. For this change, the important two changes are rate and direction of the gray, and they are expressed by the magnitude and direction of the gradient vector. Edge detection operator checks the neighborhood of each pixel and quantifies the rate of change of gray, including the determining of the direction usually. There are several kinds to be used, and the most of them are the method seeking convolution based on the directional derivative mask.