Local Energy Pattern for Texture Classification Using Self-Adaptive Quantization Thresholds

摘要: Local energy pattern, a statistical histogram-based representation, is proposed for texture classification. First, we use normalized local-oriented energies to generate local feature vectors, which describe the structures distinctively and are less sensitive imaging conditions. Then, each vector quantized by self-adaptive quantization thresholds determined in learning stage using histogram specification, transformed number N-nary coding, helps preserve more structure information during quantization. Finally, frequency used as representation feature. The performance benchmarked material categorization on KTH-TIPS KTH-TIPS2-a databases. Our method compared with typical approaches, such basic image features, binary pattern (LBP), ternary completed LBP, Weber descriptor, VZ algorithms (VZ-MR8 VZ-Joint). results show that our superior other methods database, achieving competitive database. Furthermore, extend from static dynamic texture, achieve favorable recognition University of California at Los Angeles (UCLA)