Research On Hand Gesture Segmentation And Recognition Based On Super Pixel GrabCut In Complex Background

With the development of computer technology and the popularity of intelligent equipmen,Computer interaction is very important in people's daily life.People began to pursue a kind of interacting which is more natural,more intelligent and in line with human communication habits.The information communication based on gesture is natural and convenient,and noncontact interaction can be achieved,it can meet the new requirement of man-machine interaction.Therefore,hand gesture recognition technology has been a hot spot in the field of human-computer interaction in recent years.At present,the hand gesture recognition technology has made great progress,but there are still some difficulties in the practical application as follows:(1)In practice,the background of the gesture image often contains non-gesture regions with complex shapes and various colors,as well as interference of human face and the color of skin,at the same time,changes in light make it difficult for skin models to adapt to different skin colors.(2)Due to simple geometric structure and limited features,less information can be extracted.At the same time,gestures have multiple degrees of freedom and there are differences in skin color and size of hands,therefore,it is difficult to extract gesture features.In view of the problems existing in the segmentation and recognition methods of gestures,the main contributions and innovationsof this paper include as follows:(1)In the aspect of gesture segmentation,based on the histogram contrast saliency detection model and the color distribution clustering feature of gesture,a skin Color-Contrast model is proposed.According to the regional characteristics of gesture,super-pixel segmentation is introduced and a super-pixel contrast model is proposed.In order to further optimize the segmentation of edge details,the GrabCut model is introduced.In view of the shortcomings that users need to calibrate the foreground area,skin color detection is used to automatically calibrate the approximate foreground area.The algorithm is slow due to iteration.Super-pixels are used instead of pixels,and the Gibbs energy model is improved by combining edge features.Experiments show that the model is better than the original GrabCut model on segmentational effect.(2)In the aspect of gesture segmentation,a feature extraction method based on combination invariant normalization for gesture fusion is proposed.Since gestures are deformed in threedimensional space,considering their rotation in space,affine invariant moments and Fourier descriptors are introduced on the basis of Hu moments for feature fusion.Finally,experiments show that the recognition rate of this method is significantly higher than that of a single feature.