Improving Color Image Binary Segmentation Using Nonnegative Matrix Factorization

Image segmentation is the process of dividing a digital image into multiple segments or regions, each of which represents a different object or part of the image or shares certain visual characteristics. The goal of image segmentation is to simplify the representation of an image into something that is more meaningful and easier to analyze. Typically, segmentation creates histograms indicating edges and boundaries that use local information such as color. The color space used determines the segmentation results: while RGB is the most fundamental and widely used color space, other color spaces can also be adopted for segmentation to emphasize different aspects of an image. No single color space, however, is capable of producing satisfactory results for all types of color image segmentation tasks. Nonnegative Matrix Factorization (NMF) is a powerful data reduction and exploration tool efficiently used to extract features that provide part-based representation for a wide range of data types. In this paper, we propose an NMF-based method for integrating different color space information from an image and then extracting from it an “optimal color space representation” which is used to improve the results of a threshold segmentation algorithm. The proposed approach uses NMF on histograms of different color spaces to generate a meta-histogram representation of the given image, which is then binary segmented. Experimental results showed that the features extracted from the proposed NMF-based approach are able to improve segmentation outcomes obtained by conventional segmentation algorithms like Otsu method.