A New Input Pattern of Color Distribution Information for Reduced-Reference Image Quality Assessment via Circular Extreme Learning Machine

The Image Quality Assessment system based on Color Distribution Information (IQA-CDI) uses descriptors based on the color correlogram in analyzing the distortion types and quality score, for Reduced-Reference (RR) data. IQA-CDI with RR data can predict the perceived image quality scores for real-time digital broadcasting. IQA-CDI is supported image quality prediction by the ensemble of learning machines. However, using the ensemble of learning machine in IQA-CDI may spend much time to data training process that do not suitable for real-time situations. Therefore, our research aims to decrease data training time of IQA-CDI by adapting the input features pattern for reducing the ensemble size. The experimental result shows that a new input features pattern and the reducing ensemble size can reduce processing time, while has the performance comparable to original IQA-CDI.