Approaches To Face And Skin Detection

Until now, human face detection has become an important question for face, expression and gesture recognition. It is necessary of face detection in many ways. In modern science, face detection is usually to locate the face region and recognize it in an image for further processing. It has numerous applications like: face tracking in video sequences, recognition or identification of faces, gender classification and biometric identification. To approach the face detection, we use the several ways like: skin texture, facial shape, and facial features and so on. And sometimes we need to use the combination of them all. There are many approaches to face detection. The majority of them is application of window-based classifier. It will recognize the human faces by translating a window on the entire image. But the disadvantage of detection method is not independent by illumination conditions, scale variation or pose variation. There are another 2 main approaches are applied for face detection. The first approach is image-based. For this approach, there is a pattern recognition problem treated and it is finished by many methods.

The second approach is feature-based. Invariant features of the face are used in this approach. For the skin colour, eyes, eyebrows, nose and mouth will be regarded as a feature for face detection to approach the result. To achieve the face detection, there are several a complicated and challenging task to deal with. Because of the complexity of the background, light variation and several objective factors in real situation, we need to choose the effective and applicable method for each problem we mentioned. The detected faces for recognition could be divided into different sizes, different poses and expressions, within different unconstrained lighting conditions by using skin color method. The skin pixels are being detected by skin model. The part of “face” or “not face” is recognized by the geometrical analysis and information analysis for the skin regions in the images. Because all the faces of human being have elliptical shape, the geometrical analysis can be used to deal with image. And the amount of skin region in the detected elliptical shape is sample base of the information analysis. This classification technique has the same performance for different sizes of faces also different orientation and inverted faces. Feature-based approaches build upon explicit knowledge wherein features representing a face as defined by the designer are first extracted from images.

A skin color is the main descriptor in skin detection. The computational cost of skin color is low. The skin color of skin detection is invariant to position, especially scaling. It also covers only a small part of the whole color model, which means that any presence of this kind of colored pixels in an image will be related to presence of a skin. Skin color property is the recognition for the detection of the skin colour of images. It is also depending on the colour space which is used. More color spaces used can increase the accuracy of skin detection. The common color spaces used in skin detection are RGB, normally we use the RGB in detection for the common color spaces, such as YCbCr, HSV, CIEL*a*b and TSL. The RGB color space is a cube in a 3D coordinate system. There are three additive primary colors, Red, Green and Blue in this system. They are varying from 0 to 1 on the three axes: x- axis, y-axis and z-axis. All the colors will be contained in this system. Any color will be obtained by combinations of the three colors. The HSV color space is formed by three color components: H means the hue component to define the color in the system. S means the saturation component to describes the purity of that color. V means the value component to define the intensity of color.

There is an abstraction of lighting conditions obtained by using the H and S components. YCbCr is the main color space used to represent digital video information. The brightness and two-colors difference signals are representation of a color within the YCbCr color space. Y means brightness component to represent the quality of the light. It is also the result of computation of weighted sum of RGB values. Cb means the computation of the difference between the blue component and a reference value. Cr means the computation of the difference between the red component and a reference value. Feature-based approaches is dependent by the features shown on a face region which is defined by the designer. And the face detection is achieved by recognizing the selected region in the image with a certain degree of confidence. The features extracted from the face region in an image are regarded as a face. Representative features are eigen features, Haar-like features, and edge features.