Integrated content-aware image retargeting system.
详细信息
文摘
In recent years,image retargeting has been an active research direction due to the rapid growth of mobile devices and display screens with different resolutions and aspect ratios. Traditional retargeting approaches e.g. cropping and uniform scaling) generally fail to produce satisfactory results in most cases. Although cropping maintains the aspect ratio of objects within the image,it removes some regions from the image. On the other hand,uniform scaling maintains all images regions at the expense of stretching the image in one direction aspect ratio distortion). To address these shortcomings,various content-aware image retargeting approaches have been proposed which aim at resizing an image while taking its contents into consideration to preserve important regions and,at the same time,minimize distortions. All these approaches are based on two major stages. Initially,an importance map is computed to reflect the importance of each pixel within the image. Then,a retargeting stage resizes the input image and minimizes distortions in the final retargeted image by employing the computed importance map. The objective of this research is to investigate and compare the ongoing research activities on image retargeting,and to propose a new integrated content-aware image retargeting system that outperforms other existing approaches. To achieve this goal,a solid importance map that incorporates both low level and high level salient features is a corner stone. At the beginning,an improved context-aware saliency detection approach is proposed for detecting low level saliency. While,the proposed approach produces results that are visually comparable to existing approaches,it significantly reduces the computational time. Then,various high level salient features e.g. human faces,edges,text,etc.) can be incorporated to compute the final importance map. In our research,we focused our attention to detect human faces. Viola-Jones face detector was improved by adding a skin color detector as a post-processing stage,to reduce the number of false positives. After computing the importance map,a new content-aware image retargeting scheme,called non-uniform image scaling,is proposed as the retargeting stage. Experimental study over a set of 86 benchmark images demonstrated that our proposed approach overcomes most of the drawbacks of existing approaches.