| Summary: | 碩士 === 義守大學 === 資訊管理學系碩士班 === 96 === We develop and improve an algorithm in order to detect the faces and recognize theses identity in daily life images in the varied background. We use less-dimension vectors to reduce images complexity and improving interference with noise in images,increasing ability of face detection and recognition.
The system of face detection and recognition is divided into three stages: face detection, face location, and face recognition. In the first stage, we use a fuzzy Gaussian classifier and a face feature extracting neural network to detecting faces in image. In this stage, we hope to divide images to face images and non-face images roughly by fuzzy Gaussian classifier. We compute the fuzzy Gaussian parameters of input images, and then accumulate the square errors of Gaussian parameters between training patterns to exclude the most part of non-face image. Next, we feed the passed images to the feature extracting neural network for detecting faces accurately. In the face location stage, we use Gaussian spread method to remove some fault detections in the previous detecting stage and locate the faces in images. In the last stage, we use a fuzzy c-means and a framework of parallel neural networks to recognize the faces that located in the previous stage. The fuzzy c-means can classify each input image to some clusters and activate their small-scale parallel neural networks corresponsivelyto recognize the input images.
Our algorithm can reduce the dimension of images, and eliminate a great deal of non-face images by classifier. Therefore, we can decrease the training time and recognition efficiently. Further, we can promote the detection and recognition ability of complex face images accurately.
|
|---|
