Dominant run-length method for image classification

Abstract

In this paper, we develop a new run-length texture feature extraction algorithm that significantly improves image classification accuracy over traditional techniques. By directly using part or all of the run-length matrix as a feature vector, much of the texture information is preserved. This approach is made possible by the introduction of a new multi-level dominant eigenvector estimation algorithm. It reduces the computational complexity of the Karhunen-Loeve Transform by several orders of magnitude. Combined with the Bhattacharya distance measure, they form an efficient feature selection algorithm. The advantage of this approach is demonstrated experimentally by the classification of two independent texture data sets. Perfect classification is achieved on the first data set of eight Brodatz textures. The 97% classification accuracy on the second data set of sixteen Vistex images further confirms the effectiveness of the algorithm. Based on the observation that most texture information is contained in the first few columns of the run-length matrix, especially in the first column, we develop a new fast, parallel run-length matrix computation scheme. Comparisons with the co-occurrence and wavelet methods demonstrate that the run-length matrices contain great discriminatory information and that a method of extracting such information is of paramount importance to successful classification.