Visual attention model analysis and benchmarking

Abstract

There are various existing saliency models available for performing the detection of salient regions given a set of image data. But the performance of these saliency models varies with different sets of image data used. Consequently, this project seeks to analyze the performance of the saliency algorithms at detecting the salient regions using a standardized collection of image test data. A total of five saliency models are selected for analysis and three image datasets are used to perform the experiment. The output saliency maps generated by the respective algorithms will be analyzed based on the qualitative analysis and quantitative analysis approaches. Additionally, MATLAB scripts are written to assist in automating the process of batch operations to produce the results for ease of analysis. The findings are then consolidated and suggestions for improvement to the research efforts are made.