Wavelet Domain Based Saliency Map Computation and Evaluation Using Novel Quadrant Based Approach

Abstract: Saliency map is an important and very useful representation of visually important parts of the image. It highlights the salient parts of the image by limiting the visual information coming to the human visual system (HVS). Many models have been proposed in the literature for computing the saliency maps. All of them are having their own pros and cons. No model is perfect. But it has been analyzed that instead of looking for perfect saliency model one should use the concept of multiple models each having its own advantages. From the entire available models one should choose the saliency model which is most appropriate for the application under scan. Further rather than going for high contrast or full resolution saliency maps one should go for saliency maps with acceptable resolution, giving well defined boundaries and highlighting bigger as well as smaller objects of importance. This is because saliency maps generated are generally not the final output to evaluate, rather the application, for which computed saliency map has to be used, is being evaluated. So increasing the performance of application should be more emphasized rather improving the visual quality of saliency map. Keeping this concept in mind we have proposed wavelet domain based saliency using all four components of wavelet decomposing. All four components have been used to ensure that all visually important objects, whether small or large, are highlighted with acceptable resolution. Finally this work has been evaluated on various parameters and it has been found that proposed method is good for saliency detection as compared to other state-of-art methods. An evaluation method for measuring effectiveness of the method for locating true salient region has also been proposed in this work.