| Summary: | 碩士 === 國立中央大學 === 資訊工程研究所 === 97 === Recently, gender recognition is an important and interesting research issue in the area of pattern recognition. Its purpose is to recognize the gender of an unknown person which can be applied to ensure the secure activity in gender-restricted areas, such as lady’s room. Moreover, it can provide more detail statistical information for decision making in people counting application.
Most of traditional gender recognition methods use contour-based features, such as gait energy image (GEI), which perform well only in the view angle of 90 degree. To remove the restriction, we present a texture-based gender recognition method by using local binary patterns (LBP) in this thesis. The difference between the clothing and shapes of males and females can be successfully extracted and discriminated by LBP.
In our work, the LBP histograms are firstly extracted from the foreground of inputting video sequences and concatenate them into a single vector including the LBP histograms from the whole body, upper body without skin color, and lower body without skin color. The classifier that we adopt is support vector machine (SVM) in discriminating gender. Experimental results demonstrate that the proposed texture-based gender recognition method is more insensitive to view angles than GEI. The noticeable merit of our method is that we can classify human gender by using only one single image. Moreover, the extraction of LBP features needs much less time than the extraction of GEI features.
|
|---|
