Intrinsic nonlinearity complicates the modeling of perceived quality of digital images, especially when using feature-based objective methods. The research described in this paper indicates that models from Computational Intelligence can predict quality and cope with multi-dimensional data characterized by complex perceptual relationships. A reduced-reference scheme exploits Support Vector Machines (SVMs) to assess the degradation in perceived image quality induced by three different distortion types: JPEG compression, white noise, and Gaussian blur. First, an objective description of the images is obtained by exploiting the co-occurrence matrix and its features; then, the SVM supports the nonlinear mapping between the objective description and the quality evaluation. Experimental results confirm the validity of the approach. © Springer-Verlag Berlin Heidelberg 2008.
|Title of host publication||18th International Conference on Artificial Neural Networks, ICANN 2008, 3-6 September 2008, Prague|
|Place of Publication||Prague|
|Publication status||Published - 2008|
|Name||Lecture Notes in Computer Science|