Saliency-based image enhancement

Abstract

Motivated by the importance of visual dominance in influencing aesthetics, we adopt a saliency-based approach for image aesthetics evaluation and enhancement. First, we use the saliency-enhanced approach to train models for <b>aesthetics class prediction and score prediction</b>. Next, we introduce <b>saliency retargeting</b>, a novel low-level image enhancement approach aimed to enhance image aesthetics by redirecting viewers? attention to the important subjects of the scene. Our score prediction model is used to maximize the aesthetics of the resulting image. Finally, we present <b>tearable image warping</b> for scene-consistent image recomposition and image retargeting. For image recomposition, our approach can produce an effect analogous to change of viewpoint without semantics violation, making it a powerful recomposition tool. For image retargeting, we perform better in reducing distortion and preserving semantic connectedness compared to state-of-art methods. Empirical evaluations demonstrate the effectiveness of the saliency retargeting and tearable image warping algorithms in enhancing image aesthetics.