The hatching facial sketch representation based on mixture model

Face image interpretation has been interesting issue on computer vision. Face image interpretation has been developed by some researchers, but the training and the testing sets use same modality, where for both training and testing sets use photograph for training and testing process. The training process consists of two stages, which are image labeling and giving sign on important features using landmark. The testing process consists of 2 main stages; the first, determining testing set direction, whether testing set direct to the right or to the left. The last multiple feature detection. To detect multiple features, we proposed combining the landmark movement based on gradient distribution and the simplification of movement directions. In this research, we use 200 photograph images as training set and 200 facial sketch images as testing set. For both training and testing sets used are face image and sketch with slope 20° until 90° respectively. The experimental results show that detection accuracy achieved is 83.03% for the right direction and 83.41% for the left direction. Our current proposed method is superior to the adaptive shape variants average method.