Earticle

Content-aware image retargeting has drawn much attention in image and vision research in recent years. However, existing methods are very difficult to ensure that the result images from retargeting achieve good visual effect on the whole, since these methods mainly focus on spatial image information. In this paper, we propose a new approach on content-aware image retargeting based on visual effect assessment. We establish an evaluation mechanism of the visual effects of retargeted images which is based on a priori statistical knowledge through studying the user's evaluation, and build the computable model of visual effect assessment of retargeted image with the help of mathematical description from Dynamic Bayesian Networks. After finishing content-aware processing and construct a three-level model of visual saliency contents for the original image, we retarget the original image into the target image by virtue of deforming image, and integrate computable model of visual effect assessment into retargeting process, so as to guide the retargeting. Finally, by steadily adjusting the size of the intermediate results from deforming image, we make this size be eventually equal to the size of the target image of retargeting, under the constraint that the result image should acquire good visual effect through optimizing the parameters of visual effect assessment. Real data have been used to test the proposed approach and very good results have been achieved, validating it.

Image retargeting aims to change the resolution and aspect ratio of an image to fit it into different devices. The key and most challenging issue for this task is how to balance the tradeoff between preserving the important contents and minimizing the visual distortions. In this paper, we present an effective image retargeting method called concepts-based mesh parameterization. Specifically, we first construct a structure-preserved mesh image. Based on this mesh representation, the image retargeting problem is formulated into a constrained image mesh parameterization problem, finding a homomorphous target mesh with the desired device size. To emphasize different concepts and minimize visual distortion, different concepts are first detected from the image, and are associated into the image mesh. Then the problem is transformed into solving a linear equations system. The target image is rendered by texture mapping. Experiments demonstrates the effectiveness of our concept-based algorithm.

Measurement of image similarity is a fundamental issue in both image processing and computer vision and is very important for image retargeting applications, which have drawn much attention in recent years. In this paper, we propose a new similarity measure with deformation detection of visual salient regions for image retargeting. According to the fact that the human visual system is sensitive to content changes as well as visual effect variations, the proposed approach combines content similarity and visual effect similarity. Firstly, a content correspondence between original and retargeted images is established, which aims to assess their content similarity. Secondly, deformation detection of visual salient regions between two images is conducted to measure their visual effect similarity. Real images have been used to test the proposed method and very good results have been achieved, validating it.

This paper analyzes the characteristics of the state-of-the-art technologys with regard to content-aware image retargeting in detail, contrasts the merits and demerits between diverse technologys, and lists the performance and effect about each method. The primary processes of content-aware image retargeting are divided into two steps: first step is the recognition of image importance-map, which is generally generated by the combination of image gradient and visual significance, indicates that human eyes generates diverse sensitivity of watching different area in image; Current papers with respect to content-aware image retargeting primarily centers on the resizing method in the second step. In the light of the traits of different ways, this article subdivided the technologys into three categories: on account of seam-carving, image-deformation and multi-operator. At last, we give the final qualitative analysis about the three algorithms mentioned above.

In recent years, image retargeting has been an active research topic due to the rapid growth of mobile devices and display screens with different resolutions and aspect ratios. To address this problem, various content-aware image-retargeting approaches have been proposed to retarget images while preserving important regions and minimizing distortions. In our research, the goal is to propose a new integrated content-aware image retargeting system that outperforms other traditional approaches. At the beginning, an improved context-aware saliency detection approach is applied for detecting low-level saliency. Then, a high level salient feature—human faces—is also incorporated using an improved Viola-Jones face detector with a skin color detector as a post-processing stage, to construct the final importance map. Finally, a non-uniform image scaling approach is employed as the retargeting stage. Experimental study over a set of 74 benchmark images demonstrated that the proposed system overcomes most of the drawbacks of other existing approaches.

본 논문에서는 영상의 중요한 정보를 유지하는 동시에 영상의 왜곡을 막으며, 영상의 크기를 줄이는 방법을 제시한다. 영상의 각 픽셀의 중요도를 측정 후, 가장 중요하지 않은 픽셀들을 제거하는데, 이 때 단순히 중요하지 않은 픽셀들을 제거하면, 이로 인해 영상의 왜곡이 생길 뿐 아니라 중요하지 않은 픽셀들이 가지고 있던 정황에 대한 정보도 잃게 된다. 이를 막기 위해 제시된 방법은 중요하지 않은 픽셀들을 제거한 후에, 그 픽셀들의 중요도를 주변으로 확산시키는 방법을 이용하여 영상의 왜곡을 막고, 영상의 정황을 보다 효과적으로 전달하며 영상 리타겟팅을 수행한다. 실험 결과는 제시된 방법의 효율성을 보여준다.

본 논문에서는 입체감을 보존하여 스테레오 영상의 크기를 조정하는 기법을 제안한다. 스테레오 영상의 디스패리티 데이터를 이용하여 좌우 영상의 중요 영역을 추출하고, 영상의 빈 공간을 인페인팅(inpainting) 기법을 통해 복원한다. 복원된 배경 영역의 크기를 선형적으로 조정하고, 중요 영역은 최대한 보존하도록 크기를 조정하여 배경 영역에 붙여 넣는다. 실험 결과 제안하는 알고리즘이 기존 이미지 크기 조정 기법보다 스테레오 영상의 중요 정보와 입체감을 왜곡하지 않고 보존함을 확인할 수 있었다.

본 논문에서는 스테레오 영상의 중요 부분과 깊이 정보를 유지하며 영상의 크기를 변경하기 위한 새로운 warping 기반의 리타겟팅 기법을 제안한다. 단일 영상을 이용하는 기존의 알고리즘들은 스테레오 영상간의 일관성을 고려하지 않기 때문에 독립적으로 리타겟팅을 수행한 결과 영상의 3 차원 정보가 왜곡 된다. 반면 제안하는 알고리즘은 스테레오 영상에 격자를 생성하고 격자의 정점 매칭을 이용해 두 영상의 일관성을 유지한다. 또한 영상 내의 중요 물체가 유지되는 성능을 높이기 위해 sparse disparity map 을 생성한다. 다양한 스테레오 영상을 이용한 실험을 통해 제안하는 알고리즘이 다른 알고리즘에 비해 중요 물체와 3 차원 정보를 잘 보존하는 것을 확인한다.