Image Segmentation of Pastel Drawings

摘要

What is examined in drawing therapy is the picture contents of a client's drawing. Aiming at computer-aided drawing therapy, this paper deals with an image processing technique for extracting the whole region which represents the contents from a pastel drawing. A handwritten pastel drawing often has a distinguishing feature of coarse textures, which make the image segmentation by simple thresholding difficult. Also, the picture contents of a drawing can be usually interpreted in various ways. Considering these points leads to a method for determining the whole region of picture contents in a pastel drawing image so that the requirements of the practice of drawing therapy can be met. In the method, first, painted pixels are obtained by thresholding. Then, the set of the pixels are fused by morphological processing, and thereby the region boundary is closed. The method has variables in each processing so that the resulting region can be varied. Experimental results demonstrate how effectively each variable performs in varying the figure of a resulting region according to the interpretation of picture contents.