18. Feature Presentation of Image Saliency Existence Based on Boundary Compactness Hypothesis

The study of salient image is related to the method for detecting the most prominent area within the image. There were numbers of methods applied in describing the image saliency, and the models’ integration with the learning techniques has been able to provide tremendous achievement on the results of salient object detection. However, the approach that is able to describe the saliency existence has never been seriously discussed, and therefore, many models are still unable to produce correct detection for the non-salient image. The non-salient image is a type of image that does not contain any important information. This paper presents a method that can describe the saliency existence of images based on the boundary compactness hypothesis with fact that the compactness of a non-salient image is spatially distributed as being referred to its boundary compared to the salient image. The saliency features were extracted from the image background measurement that consists of boundary contrast compactness and boundary spatial distribution compactness. These compactness components were computed for all image’s superpixel patches and compared with its boundary patches. As these features were computed in the spatial domain, the fast Fourier transform is applied to obtain the saliency features in the frequency domain. Experimental results show that the proposed approach achieve the highest mean difference ratio of 21.65 compared to the state-of-the-art approaches in putting distinct value to identify the saliency existence.