Research On Method Of Model Construction For Feature Binding Of Color Image Based On PCNN

The research of this paper is an important component of the National Natural Science Foundation "brain cognitive process and model research for feature binding of image color and shape"(ID:61070077), aiming at the research on the method of model construction for feature binding of color image.As a research hotpot and difficulty in the field of computer vision, image recognition that has wide application background has long attracted the attention of many scholars. This paper focuses on the research of construction method for the feature binding neural network model of image color and shape, which provides the theoretical basis and the reference model for computer image recognition.Aiming at the construction method of feature binding model for color image, the paper does a lot of work that as follows.Firstly, this paper analysis of the deficiencies in the color image processing of the PCNN model based on strong, and constructs a feature binding PCNN model base on vector (VFB-PCNN). In this model, PCNN model of gray scalar space expands to VFB-PCNN model of color space. And the model achieves feature separation as well as feature binding of color and shape features by using the time matrix, while discovers the optimal iteration times, sets up an automatic binding model of color image. The experimental results indicate that VFB-PCNN model can well solve the automatic feature binding problem of color image.Secondly, the VFB-PCNN model can well solve the automatic feature binding problem of color image, but it cannot identify all of the colors in the color space. Therefore, this paper advances Double Space Vector Features Bundled PCNN Model (DVFB-PCNN). The model successfully solved the problem that the VFB-PCNN model cannot separate all the colors in the color space by utilizing the method of combining the RGB color space and HIS color space, and it has a good robustness.The present model can well solve the binding problem of color image, but the constructed model is only based on the basic features of the image while not make full use of visual processing mechanism. The paper puts forward that if the visual processing mechanism is introduced into the model to construct the binding model, it will be more close to the human brain processing of color and image as well as more objectively reflection the actual information processing.