JPEG-coded Image Quality Assessment Based on GAP-RBF Neural Network
Huaibo Song · Journal of Chinese Computer Systems · 2013
A novel no-reference method to assess the quality of JPEG-coded image using Growing and Pruning radial basis function(GAP-RBF) neural network is proposed in this paper.GAP-RBF neural network is a sequential learning algorithm neural network.The sequential learning algorithm requires minimal computational effort and memory because of its capability to learn new samples without retraining the past learning.The features for predicting the perceived image quality are extracted by considering human visual perceiving characteristic.From computation of the block-based Discrete Cosine Transform,each 8×8 image block is categorized into either ′texture′ or ′smooth′.For the latter,we further detect its edges which is considered as artificial discontinuities across the block boundaries.Thus image features are the statistics information of the artificial edges.The features are connected to the input layer of GAP-RBF neural network so that the functional relationship between features and subjective test scores is modeled by GAP-RBF neural network.Experimental results prove that objective score and mean opinion score have good consistency.