Pulse Coupled Neural Network Image Segmentation And Face Detection

PCNN has rapidly developed as a new tape of artificial neural network since 90s of the 20th century. PCNN's neuron model is an artificial neural model obtained by simulating visual neural activity, so this neural network model has biology research background. Due to in the environment of weak connection, PCNN has the features as scale invariance, rotation invariant, signal strength invariance, signal distortion invariant and especially it can change two-dimensional spatial variables into one-dimensional time series at the same time processing images, PCNN is very suitable for image processing environment. The applications relate to various aspects, such as image segmentation, image target recognition, moving object recognition, artificial life and decision optimization, therefore, PCNN has aroused the attention of scholars (at home and abroad)and been widely used in image processing. This paper used PCNN model parameters estimation and image segmentation based on PCNN and color of skin, face detection has done some research.In the image segmentation, through the analysis of the working mechanism of PCNN, we use simplified PCNN to process image segmentation. However, in related research literature, the model parameters were determined by the experimental artificially more. This paper uses wavelet analysis to process multilayer decomposition of the image, and then, with multilayer decomposition of low-frequency coefficients reconstruction images as links to weight W parameters estimation of the model, reoccupy an optimal threshold value method to estimate threshold theta 0, finally network computing iteration N is determined by maximum relevant criterion, image segmentation is successfully achieved. Experimental results compared with relevant literature, the details of image segmentation make great improvement, the quality of image segmentation also gets bigger improved.In face detection, aiming in simple skin detection easy to mistake by with color features detected, we design a detection method of skin model based on PCNN. For color image preprocessing first proposed the white balance algorithm to process light compensation, this method has a good compensation error caused by external light for skin segmentation, thus it makes for skin regional segmentation. Then the image is transformed from RGB color space to YCbCr color space, and will process skin image segmentation in YCbCr space, and will get some sub-block of the color of skin area. Finally using PCNN to extract all sub-block ignition timing sequence, calculate target face and various sub-block difference degree in being tested image, The smallest difference degree is one with the artificial same individual target face. This method can be used for different focal length face detection. To algorithms mentioned above, a large number of experimental tests are done, the results show that the effect is good.