New Threshold Setting Method for the Extraction of Facial Areas and the Recognition of Facial Expressions

A new method for setting a threshold value on a quasi-chroma histogram to extract facial areas based on the methods used in forecasting stock market features, such as dead cross and golden cross, is proposed. In general, when a threshold value is set to isolate objects from a background, the histogram of an image is taken and examined. However, the shape of the histogram constructed from an original image is generally very jagged. Before setting the threshold, a smoothing operation is applied using a mask with the interval n. Here, the problem is in deciding the interval n and the iteration time. In general, the required interval n differs from picture to picture. Therefore, choosing the interval n is quite a delicate process. However, in this paper two feature points, dead cross and golden cross , which are generally used to decide the timing of selling or buying stocks, are adopted to set the threshold value for a quasi-chroma element image. Based on the hue element, the skin-color area is first isolated. A facial area is then extracted from the skin-color area based on a quasi-chroma histogram constructed from the skin-color area. In this step, the threshold value is set at the point where the frequency is at a minimum between the two feature points. The evaluation of the repetition time of smoothing is very important. An adequate number of iteration and interval n are also examined through experiments. In the final part of this paper, the recognition of expressions is presented