Adaptive Energy Function for Active Net.
Kazuyoshi Yoshino, Mio Maki, Toshio Kawashima, Yoshinao Aoki · 1994
This paper proposes an algorithm to extract a target region from a color image. The algorithm utilizes the active net for stable target extraction. In order to extend the original active net for color image analysis, the image energy function of the net is defined by the function, called the Ratio Histogram, of the histogram backprojection method. Since ratio histogram is computed from both color histograms of a target and an input image, the adaptive image energy function for the input image can be created. Then, since our proposed method uses target colors as the index of the target, the net can extract a target regardless of the target size and of its pose. Additionally, we refine the proposed image energy dynamically during iterations of the net to improve the net convergency. Experimental results for color images taken by a video camera show the validity of our approaches. 1