IEICE Transactions on Information and Systems
Online ISSN : 1745-1361
Print ISSN : 0916-8532
Regular Section
Minimum Mean Absolute Error Predictors for Lossless Image Coding
Yoshihiko HASHIDUMEYoshitaka MORIKAWAShuichi MAKI
Author information
JOURNAL FREE ACCESS

2008 Volume E91.D Issue 6 Pages 1783-1792

Details
Abstract

In this paper, we investigate minimum mean absolute error (mmae) predictors for lossless image coding. In some prediction-based lossless image coding systems, coding performance depends largely on the efficiency of predictors. In this case, minimum mean square error (mmse) predictors are often used. Generally speaking, these predictors have a problem that outliers departing very far from a regression line are conspicuous enough to obscure inliers. That is, in image compression, large prediction errors near edges cause the degradation of the prediction accuracy of flat areas. On the other hand, mmae predictors are less sensitive to edges and provide more accurate prediction for flat areas than mmse predictors. At the same time, the prediction accuracy of edge areas is brought down. However, the entropy of the prediction errors based on mmae predictors is reduced compared with that of mmse predictors because general images mainly consist of flat areas. In this study, we adopt the Laplacian and the Gaussian function models for prediction errors based on mmae and mmse predictors, respectively, and show that mmae predictors outperform conventional mmse-based predictors including weighted mmse predictors in terms of coding performance.

Content from these authors
© 2008 The Institute of Electronics, Information and Communication Engineers
Previous article Next article
feedback
Top