Research On Blind Separation Of Image Sources Algorithms Robust To Rigid Transformation

At present, in the increasingly automatic printing industry production, more and more traditional product surface quality detections based on eye recognition are depend on automatic visual system. In the printing industry production, one of the most important and critical step is modeling. However, in the process of the printing, there may be exist some rigid transformation like rotation and translation or the other noise which will bring a potential quality hazard to the approximation of the modeling results. In order to improve the quality of modeling, according to blind sources separation and convex optimization, this paper tries to design a novel blind separation of image sources robust to rigid transformation to achieve the more approximate modeling results, improve the modeling quality and decrease the error detection rate of the surface printing products. This paper mainly includes the following aspects:First, this article firstly discusses the background and theoretical foundation of the BSS in a systematic way, and then abstracts the practical problems of the printing products surface quality detection modeling system for the mathematical model of the BSS, finally explores the new BSS method to solve this problem. This paper illustrates the theoretical basis of the BSS (Blind Sources Separation) which can be applied to visual detection system.Second, this paper explains the mathematic analysis of the BSS applied to the printing products surface quality detection modeling. According to the classic BSS methods and the demands of the printing products surface quality detection modeling system, we point a suitable BSS method called CAMNS (Convex Analysis of Mixtures of Non-negative Sources) and give a rigorous mathematical foundation for the blind separation of image sources algorithms robust to rigid transformation. Experimental evidence shows that although the CAMNS algorithm can solve a part of the new problem, we still need to improve the existing BSS algorithm to resist the influence of the rigid transformation in the printing products surface quality detection modeling system.Third, the innovation of this thesis is to propose a new research on blind separation of image sources algorithms robust to rigid transformation. This method firstly proposed a tentative plan: uses FMT (Fourier-Mellin Transform) to preprocess the observed images based on the CAMNS algorithm, and then the rigid transformation invariant factors can be determined. But according to the experimental results and certificate, we find that the FMT can not settle the problem still. Therefore we redesign a novel CAMNS algorithm based on geometric transform. This new method firstly preprocesses the observed images, and then extracts the rigid transformation invariant factors. According to the special assumption called local dominance, the issue of blind image sources separation which is influenced by translation and rotation turns into a solvable convex optimization, through which the mixing matrix can be estimated. Finally solve the mixing equation group to get the image sources. Experimental results demonstrate that this novel algorithm is quite effective for blind separation of image sources robust to rigid transformation.