This thesis proposes a method for automatic discrimination of the gender of a person in a color image. The person in a color image is extracted by subtracting the input image from the background image. Features such as the line of a skirt, hair style, clothes color and lip color are extracted, the mean value of each extracted feature is stored, then, the vector of the feature is estimated with a linear transformation according to the reliability of features. For learning the distributions of features, many color images whose gender are known are classified into two clusters by using K-means algorithm. The male feature vector model and the female feature vector model are obtained from these clusters. The gender discrimination is performed by comparing the feature vector of unknown input image with each model. The algorithm was applied to 55 test images, and about 90% of the gender was discriminated correctly. The experimented results show the effectiveness of the proposed method.
To Home Page of Chihara-lab