Image categorisation based on the spatiochromatic information

Abstract

An imaging workflow to categorise images into business graphics and pictorial images based on spatiochromatic information is demonstrated. Studies on three computational characteristics (lightness frequency distribution, average chroma distribution, and Fourier spectrum) show that differences between business graphics and pictorial images can be discriminated mathematically. To enhance the performance of the image categorisation workflow, a multilayer perceptron neural network was designed and experimentally evaluated, and which performed with an accuracy of 95% using a database consisting of 57 randomly selected images.