Reliable Research Of Edge Detection Method

People mainly through Image to understand the world,and through the edge of image to understand the position and the contour of the object in image, then further to image analysis and understanding, so that the edge is an important feature in the field of image processing, edge detection method research has become one of the hotspot and difficulty in the research of image processing. Image gray value of mutation generate edge, so all of the edge detection algorithm is based on a first-order differential operator or second order differential operators, this paper proposes a continuous curved edge detection algorithm based on DCT.Probably because of multiple factors the edge can’t be accurately detected, such as in image capture process image subject to noise interference, natural scene projection and the intensity of illumination, even in the process of image transmission, it also affected by noise, so edge detection is an essential step in denoising, one of the most commonly used denoising method is Gauss filter,but Gauss filtering is difficult to balance the image denoising and edge preservation, the reason is noise,texture and edge all belongs to the high frequency component of image, so Gauss filter as the low pass filter when it remove the noise it blurring the edge and texture details in a certain extent at the same time, and the inhibitory effect of Gauss filter for impulse noise is very poor, it is easy to take impulse noise as edge; in addition,the parameter to the control of the degree of smoothness is predefined. The bigger of the parameter,the better effect of denoise performance,but it also easier to lose some details of the information, The smaller of the parameter,the effect of keep image edge details is better, but it can`t reach the ideal denoising effect, so in order to obtain the better denoising effectively and maintain the edge and texture at the same time, to compensate for the fuzzy of edge and texture in denoise process, the most commonly approach used is use histogram equalization and histogram horse paired for image enhancement processing, but the two enhancement processing is enhance the whole image,it can`t enhance edge and texture information only. In this paper we present a image edge detection method based on DCT,witch transform the image information from spatial domain to the frequency domain, after obtain the DCT coefficients, as we all know the image information energy will be concentrated in the low frequency part, texture region of DCT coefficients are smaller in magnitude; edge region of the DCT coefficient magnitudes larger, we use the cube of the originalcoefficient divide the sum of it square with a constant to updated these coefficients so that to achieve the denoising effect.Although since from the fifties of the 20 th century when proposed edge detection technology,people have never stopped the study on the edge detection algorithm, but because of the importance of edge detection and the research relates to the depth and difficulty, therefor it is still not a widely accepted universal edge detection algorithm so far. The LOG operator is considered optimal Second-Order Differential Operator, it combine Gauss filter with Laplace operator, the above we proposed using DCT denoising instead of Gauss filter, here we use a new method to replace the Laplace operator, the reason for this is that after the implementation of denoising through updating the DCT parameters, we use these new parameters(decimal Series)conduct IDCT, and compute second order partial derivative of the IDCT formula, we take the horizontal direction and vertical direction of second order partial derivative as the sum second order partial derivative,because IDCT is the superposition of a series of cosine function, it is a continuous function,after compute the sum of second order partial derivative,we find that it still a the superposition of a series of cosine function, so it is continuous, we will regard it as a continuous surface, the second order partial derivative sum of each pixel is sampling on the continuous surface, the result we get is a decimal not an integer, so Laplace second-order partial derivative which through zero crossing to locate the edge here does not apply.For each pixel we use it’s 2x2 neighborhood to locate edge,if the sum of second order partial derivative is all positive or negative we think there no edge point in this neighborhood, else we take the pixel as edge point which absolute value of the pixel’s second order partial derivative sum is minimum, cycling the entire image for all the edge points.