The Study Of Feature Extraction Based On Complexion Title

Face feature detection in computer vision, computer graphics, a hotspot and difficult. In this paper, facial feature detection techniques are studied, proposed a new constraint-based hierarchical face detection system framework. Facial features localization in face recognition, 3D virtual face synthesis, facial expression analysis and synthesis has a very important role, is the current computer vision, computer graphics, pattern recognition and other fields of research.Currently, there are many algorithms already proposed and the model, however, face complexity and because face images imaging environment, has not yet a general algorithm or models can solve this Yanjiunanti.Face a positive image contains many significant feature of the angles from the two organs of the feature extraction: a geometric relationship between facial structure by prior knowledge of structure-based method of extracting the knowledge level of Facial major organ characteristics, such as nose, eyes, eyebrows and mouth and other features. When prior knowledge of the structure of the object, but the size, location and the center is uncertain, deformable template is an effective method of positioning. Another is the level of the signal feature extraction organs will face as a high dimensional image space in signal detection problems. Formed using principal component analysis, "Eigenfaces" as the matching features, the face image size, position, illumination, the case can not effectively extract facial feature.In recent years, facial expression and facial feature recognition has attracted the attention of many researchers. In daily life, facial feature provides us with a lot of rich information. For the face of their search in the archives, identity testing, and video communications for the great prospect, much concern researchers as a very active area of research and study. Many people assume that the face of the work is the image of the human face location have been determined, but in order to design the automatic face recognition system, fast and efficient face detection is the need to assess the main criteria. Face has many advantages, such as non-invasive, relatively Wen Ding, easy access, the only Deng, making it the biometric technology areas. Therefore, face recognition is the most natural means of identification in the machine, a biometrics research hotspots. However, as is the plasticity deformed face, the face will change with age, combined with the image generated by the range, scale, rotation and illumination and other related factors, the performance of the model is relatively complex, so the employer face more difficult to authenticate. Face to use to pattern recognition and image processing knowledge, and possibly in the computer vision and neural networks. Therefore, face recognition is a very difficult task, even though the field of face recognition has been some progress now, but only in the research of a more preliminary stage.In the human face, the skin color is undoubtedly the largest proportion, while the color changes relative to the facial expression, facial changes are more stable posture. At the same time can be better distinguished from most of the background color the color of objects, so color features a face detection is the most commonly used features, but also often as an image preprocessing method.Through a series of image processing, to extract the general location of the face, mainly to facilitate facial feature extraction, facial feature to reduce the scanned area, thus accelerating the rate of facial feature extraction. As the image color information often light, the color deviation acquisition equipment and other factors leading to an overall color to a move that we often see the colder, photos yellowish and so on. In order to facilitate image processing to offset this there is to see the whole image, color deviation, the brightness of the image pixels arranged according to low and high brightness of the pixels taken. Here we take the top 5% of the pixels for the "reference white", but also about colors R, G, B component value adjustment 255. To be compensated image light.Thanks to the image contains a lot of discrete information, the positioning will face cause some interference, so we use dilation and erosion to remove discrete information. Because non-face relatively small size of the area, it can be removed under this point. Prime Minister scanning the entire image, if you encounter a white point, judge, and the white point directly or indirectly linked to the number of white spots are big enough to do, select the largest connected area, then this region is a human face, the rest should be removed. In determining this area, they face feature extraction areas will be further narrowed, which reduces the processing pixel, to speed up the image processing speed, so to can greatly reduce the computation time. Organ in the human face plays an important role in the eye, which is relative to the face as other organs, as long as the eye can be precisely positioned, it can be the eyes and eyebrows, nose, mouth and other organs of the distribution of potential relationship between a more accurate positioning .In the human face image recognition, regardless of the geometric features of images, or images of algebraic features, are required to preprocess the image. As the center distance between two facial expressions change by the light or the least, so the eyes face facial feature extraction often as the focus.Eye brightness value based on the image to determine the pixel brightness, can use the YCrCb space, the brightness value Y, where RGB color space directly in the direct calculation of brightness. Locate the eye region, got two more regions, in order to further pinpoint the face, but also to locate the centerIn positioning the eye, you can set below the mouth of its candidate characteristic area, looking for color and larger in non-regional candidate as the mouth, and then set the color of the upper and lower limits conditions to be extracted on the eyes. Similarly in the treatment of such eye location operation in YCrCb color space is Cr, Cb component operation.In this paper based on color, color space conversion, YcrCb matched facial feature location algorithm can balance the speed and image recognition to detect the stability of contradictions.Used in the feature location detection from coarse to fine strategy, the face image to extract color, skin color regions through analysis of the orthogonal transformation and determine the candidate region; and then, in the candidate face area, its color, brightness matching According to the geometric characteristics of the human eye location method used for coarse filtration; Finally, determine the characteristics of the region through the expansion of corrosion, by calculating the level of complexity of candidate block filter to get the final result.In the experiment, also RGB color space extraction method, YCrCb color space, color extraction method combined with the YCrCb color space based on skin color extraction algorithm were compared and analyzed. Found by extraction of YCrCb color space than RGB color space is better extraction can not only distinguish between good and non-color point color points, but also a better extraction of color points.