Image content classification using a block Kolmogorov complexity measure

Image content classification is a very important step in document image analysis and understanding, and page-segmentation-based document image compression. In this paper, we present an new approach to classifying image content using block Kolmogorov complexity (KC) measures. A binarized two-dimensional image is first partitioned into blocks and each block image is converted into a one-dimensional binary sequence using either horizontal or vertical scanning. The block complexities are then computed over the obtained binary sequences. An image is classified into one of two categories, textual or pictorial images, using two fuzzy rules with the mean value and the standard deviation of block complexities. Experimental results on eight Chinese/English textual images of different fonts and eight different pictorial images show that our approach is reliable in discriminating these two types of images. Moreover, the performance of our method, where a training process is not required, is comparable to that of a neural network technique.