An Integrated saliency detection model based on weighted global entropy

Visual saliency detection plays an important role in computer vision applications. This paper proposes an integrated saliency detection model based on weighted global entropy. Firstly, in the calculation of element saliency, new definitions of element uniqueness and compactness with logarithmic responses are proposed, which can reduce the detection model´s sensitivity to object interior textures and ensure the completeness of the detected salient object. Then, a nonlinear fusion method based on weighted global entropy is proposed and applied to combine the saliency maps that are separately calculated in multi-scale Lab and RGB color spaces. Lastly, the proposed saliency detection model is compared with 12 state-of-the-art methods on a publicly available dataset. Experimental results demonstrate that our model can improve the precision, maintain a higher recall ratio, and produce saliency maps closer to human-labelled salient regions.