Application of new advanced CNN structure with adaptive thresholds to color edge detection
Shaojiang Deng,Yuan Tian,Xipeng Hu,Pengcheng Wei,Mingfu Qin
DOI: https://doi.org/10.1016/j.cnsns.2011.09.007
IF: 4.186
2012-01-01
Communications in Nonlinear Science and Numerical Simulation
Abstract:Color edge detection is much more efficient than gray scale detection when edges exist at the boundary between regions of different colors with no change in intensity. This paper presents adaptive templates, which are capable of detecting various color and intensity changes in color image. To avoid conception of multilayer proposed in literatures, modification has been done to the CNN structure. This modified structure allows a matrix C, which carries the change information of pixels, to replace the control parts in the basic CNN equation. This modification is necessary because in multilayer structure, it faces the challenge of how to represent the intrinsic relationship among each primary layer. Additionally, in order to enhance the accuracy of edge detection, adaptive detection threshold is employed. The adaptive thresholds are considered to be alterable criteria in designing matrix C. The proposed synthetic system not only avoids the problem which is engendered by multi-layers but also exploits full information of pixels themselves. Experimental results prove that the proposed method is efficient. © 2011 Elsevier B.V.