System and method for detecting colors of skin

Abstract

The invention discloses a system and a method for detecting colors of skin. The method comprises the following steps of: taking a plurality of frames of a video as a sample video and performing color space conversion on the first frame of image of the sample video; training a skin color lookup table through collecting sample pictures; for the first frame of image, performing rough skin color detection by means of the skin color lookup table; performing EM clustering on the initial skin color sample which is roughly detected previously so as to establish an initial Gaussian distribution model; performing skin color classification on the second frame of video of the sample video by using the initial Gaussian distribution model, thereby obtaining a new skin color sample; and performing EM clustering on the new skin color sample, thereby updating the initial Gaussian distribution model, and performing the same operations on the rest frames until all frames of the sample video are processed. The system and the method provided by the invention are capable of achieving the purposes of good robustness and high detection accuracy.

Description

Technical field [0001] The invention relates to a skin color detection system and method, in particular to a robust online skin color detection system and method. Background technique [0002] Currently, the common skin color detection methods are generally as follows: [0003] 1. Fixed threshold method: Due to the strong clustering of skin color, the distribution in the color space is relatively concentrated. Using mathematical expressions to clearly define the skin color range is a simple skin color modeling method. Usually YCbCr color space can be used, when the color of the pixel falls into Cr A [133, 173] and Cb A The rectangular area defined by [77,127] is considered to be a skin color pixel. [0004] 2. Parameterization method: It has a clear functional form, and can obtain corresponding parameters through training data. The training and classification speed is slow, such as using a single Gaussian model or a mixed Gaussian model to model skin color. [0005] 3. Non-parametric...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the disadvantages of low generalization performance and difficulty in adapting to complex backgrounds common in the existing skin color detection

Images