21-05-2014, 03:48 PM
TOWARDS A BIOMETRIC PURPOSE IMAGE FILTER ACCORDING TO SKIN DETECTION
TOWARDS A BIOMETRIC.doc (Size: 1.07 MB / Downloads: 27)
Abstract
In biometric recognition, sign language processing, automatic age estimation, gait interpretation and other image processing oriented studies, detecting various regions of the human body such as face, hands ect. has great importance. Furthermore with the rapid development of internet, building up the image filters is required. In this study a biometric purpose image filter according to skin analysis is mentioned. For skin analysis various color spaces like RGB, normalized RGB, HSV, YCbCr and TSL studied and transformations in these spaces are performed.
Introduction
Skin detection is an important first step in face detection, recognition, facial expression extraction, face and hand tracking and other many recognition systems. For example in face detection with neural networks (NN), the NN processes every location of the image in order to detect faces anywhere. Before this process if skin detection is applied to the image, the NN only processes skin regions, so the performance of the system increases. As well as face detection with skin analysis, if the parameters changed, regions like hair, lips can be detected. In automatic lip reading, detecting only the lip region, for gender estimation detecting head and hair information is important. Also for age estimation skin analysis for wrinkle detection is required. The skin detector can also be used as the basis for an adult image detector [1]. There is a growing industry aimed at filtering and blocking adult content from web indexes and browsers.
There have been several studies and different approaches proposed on skin color modeling and recognition. The goal of skin color detection is to build a decision rule that will discriminate between skin and non-skin pixels based on color components. In order to obtain adequate distinction between skin and non-skin regions, color transformation effectively separating luminance from chrominance is needed.
Skin detection techniques usually use color components in the color spaces, such as HSV, YCbCr, TSL or YIQ even though the input image is generally in RGB format. That is because RGB components are subject to the lighting conditions. Skin tones vary significantly within and across individuals. Furthermore the skin detection methods are insufficient when the non-skin regions closed to skin tones takes place in the image.
A wide variety of color spaces have been applied to the problem of skin color detection [2]. Which color space is the most efficient one is still a contentious question. For Shin et al. RGB and YCbCr spaces has the best performance for skin detection [3]. From Terrillon et al. study it can be observed that the TSL chrominance space is very effective for skin segmentation when using a Gaussian model. This is also true where illumination conditions vary [4]. Pietrowcew’s study it was confirmed that the best results in skin color selection are achieved for the TSL space [5]. Sun et al. reported that although skin colors of different people vary over a wide range in color space, the variation of human face color with respect to hue and saturation is much less than to brightness [6].
Color Spaces
A wide variety of color spaces have been applied to the problem of skin detection. In this study RGB, normalized RGB, HSV, TSL and YCbCr color spaces are analyzed and transformations in these spaces are performed.
RGB (Red, Green, Blue) and Normalized RGB Space
Red-green-blue space is one of the most common color spaces representing each color as an axis. The RGB components are dependent to illumination conditions. For this reason skin detection with RGB color space can be unsuccessful when the illumination conditions change. Peer et al. defined the boundaries of skin cluster in RGB color space [7]. RGB is classified as skin if the following conditions satisfied.