Face Detection In Color Image And Facial Feature Location

The research of face detection involves pattern recognition, image processing, physiology and other disciplines. It also links closely with identification methods that based on biological characteristics, and human-computer apperceive interaction fields, on the other hand face detection is the foundation and premise of face recognition. At present, algorithms and practice for face detection in simple background have been more mature. Nowadays automatic face recognition systems which are appearing on the market mostly required a forward face and simple background conditions. But as for face detection in complex background, although appearing a lot of algorithms, still failed to establish a common and efficient algorithm which cannot achieve the balance between accuracy and real-time performance.First, we introduced the background and significance of face detecting, difficulty and criteria, finally analyzed the development status of face detection in color image.Then, we analyzed the clustering of face color in three commonly used color space which including RGB. YCbCr and HSV. And we made illumination compensation on image preprocessing by using "white reference" method for the reason of color deviation when image was ingested. Then for the distributions (clustering) of face color in the color space, I choose the color space which makes a combination the YCbCr and the HSV for detecting the candidate regions of face.Next, we did a research on variety of algorithms of face detection. And we made an improvement of the traditional SNoW algorithm which was resource-intensive, inefficient and poor robustness. Though the method of local binarization mapping algorithm for the local brightness, the pixels only in 16 possible values that saved storage apace significantly. Finally we trained the SNoW classifier by making a combination of Yale faces library and Successive Mean Quantization Transform (SMQT) methods, and using winnow rules updating the training process. The resulting shows that it can detect both single and multiplayer situation with high accuracy and get rid of "human-like face" interference.Moreover, we introduced a variety of algorithms of eye location. I firstly made an improvement to the gray model of Rein-Lien Hsu, and then made a combination with the chrome model of Rein-Lien Hsu. The improvement could achieve a coarse detection eyes of weaken the interference of nose and mouth. But he results also showed that it failed to detect one of the eyes due to the deflection of light and face. Then I made use of Gabor transform for eyes detection and it showed that there was a certain y-offset due to the size of face. As for that, pursuant to the first finding of the vertical offset of the eyes to correct the eyes'position of the second finding. And the experiment shows that it can accurately make a realization of eyes detection.Finally, we also introduced a variety of algorithms of mouth detection. Then, We filtered out the candidate areas of mouth for the lip color information in YIQ color space. Next we took the mouth mapping method and got the biggest values of the function which was the mouth position of center point. The experiment shows that it can accurately make a realization of mouth detection.In the future there will have multiple information fusion for face detection, and the accuracy and speed of detection will also be improved. Static images of face detection technology research will also promote the dynamic detection technology.