Novel and more effective photometric invariant contour detector using improved color tensor

At the beginning of the paper, we discuss the inadequacies of existing contour detectors. The chief inadequacies are, in our opinion, ignoring color information and detecting only a single type of feature such as edge or corner. In order to minimize the inadequacies, we propose a novel and more effective photometric invariant contour detector using an improved color tensor. In the full paper, we explain our contour detector in some detail; in this abstract, we just add some pertinent remarks to listing the one topic of explanation. The topic is: algorithm. Its five subtopics are: improved color tensor (subtopic 1.1), connection between the improved color tensor and tensor voting (subtopic 1.2), photometric quasi-invariant model (subtopic 1.3), the simultaneous detection of edge and corner (subtopic 1.4), the flow chart of the algorithm (subtopic 1.5). In subtopic 1.1, we utilize a new type of nonlinear filter to improve the color tensor that uses the Gauss kernel and we apply tensor mathematics to reinforcing color vectors with opposite directions. In subtopic 1.2, we establish the connection between tensor voting and the improved color tensor in the nonlinear filter. In subtopic 1.3, we combine our improved color tensor with the photometric quasi-invariant model proposed by J. Weier et al in Ref. 9. In subtopic 1.4, we combine edge response with corner response to simultaneously detect edge and corner for contour description. Finally we perform experiments to verify the effectiveness of the algorithm. The experimental results, shown in Fig. 3 and Tables 1 and 2 in the full paper, demonstrate preliminarily that our method can effectively enhance the performance of contour detector.