Abstrato

Feature Fusion Based Cartoon Character Retrieval Using Semi-Msl

J.Ida princess, V.Nithya, P.Arjun

Cartoon is popular and successful media in our life, its creation is usually of high cost and labor intensive. The computer assisted systems are designed to reduce the time cost of the cartoon production and to develop novel systems that can synthesize new cartoons by reusing existing cartoon materials. For that purpose, we propose a method for cartoon character retrieval with two applications, namely content based cartoon image retrieval and cartoon clip synthesis. In order to retrieve similar cartoon characters two issues are considered. The first issue is how to represent the cartoon characters. The color histogram, hausdorff edge feature, skeleton features and curvature scale space are used to represent the color, shape and gesture of the cartoon characters. The second issue is to combine these features using semi-supervised multi-view subspace learning algorithm. This algorithm uses patch alignment framework to construct and align local patches. Aligned patches have high dimension, it is not efficient to retrieve similar character. In order to improve the efficiency, alternating optimization is utilized for reducing high to low dimensional subspace. Within that space similar characters can be retrieved. This method is useful for animators and cartoon enthusiasts to effectively create new animations.