The Application of Sparse Coding Algorithm Based on Kurtosis Criterion in Natural Image Compression

• Published in: 2009 Chinese Conference on Pattern Recognition pp. 1 - 5
• Main Authors: Li Shang, Wen-jun Huai
• Format: Conference Proceeding
• Language: Chinese; English
• Published: IEEE 01-11-2009
• Subjects: Analytical models; Discrete cosine transforms; Feature extraction; Image coding; Independent component analysis
• ISBN: 1424441994; 9781424441990
• DOI: 10.1109/CCPR.2009.5344107

This paper mainly discussed the algorithm of sparse coding based on the kurtosis criterion and its application in natural image compression. Sparse coding of natural images is in fact a transformation coding method, and it can efficiently perform extracting natural images' features and compressing images. Utilizing the kurtosis as the punitive function of sparse coding method's sparsity, it can ensure both the sparsity and independence of feature coefficients of natural images, and extract more efficiently the edge features of images. Compared with the image compression methods of standard independent component analysis (ICA) and discrete cosine transfer (DCT), the simulation results show that our method proposed excels in natural image compression.