Research On Face Detection Based On Pulse Coupled Neural Network

The Pulse Coupled Neural Network (PCNN) was experiment result which obtained to explain the synchronous burst of the neurons in the animals' visual cortex. The properties inherited in PCNN, such as linking field and the dynamic threshold, make the approximate neurons fire simultaneously, and this is very close to the natures of visual cortex of small mammals. So it has gained widely applications in image processing, such as image segmentation, edge extraction, object recognition and so on.In this thesis, through reducing some parameters we get a modified PCNN which also has the linking field and the dynamic threshold, and MPCNN is adapt to image processing. This thesis investigates the model of MPCNN, which includes its basic principles, behaviors and characteristics. According to this, the thesis introduces image entropy and brings forward an algorithm based on MPCNN and image entropy. Applying this algorithm to ORL database, detection results indicate that it is an effective algorithm.Because of object and background have obvious differences in ORL database, to color images, it not only contain brightness information, but also chroma information, otherwise, the background is different in each image, object and background is difficult to division, experiment results show that the algorithm based on max entropy is not effective to color images. Thus the thesis brings forward another algorithm which is based on the skin color algorithm and MPCNN algorithm. Then applying the algorithm to color lena image, Purdue-AR video dynamic images, which have different face expression and different light, and common color images that come from Internet. To segment images, this thesis do some processing to locate face, in addition, this thesis compares the algorithm with algorithm based on template matching, the result shows this algorithm is well suited for further recognition.