Intelligent Image Resizing Based On The Visual Attention Mechanism

With the rapid development of modern science and technology, various display devices of digital images have been produced, such as high definition digital television, notebook computer, mobile phone and so on. However, once given a digital image, whose resolution and size is constant, people always concern that how to display the image on the different devices with varied resolution and size. Consequently, image re-sizing has been one of the most hot research topic in the area of digital image processing recently.Two core requirements of image size are as follows:(1) The size of the digital image resized should match that of the display device.(2) The important content that people are concerned with in the original image should be well retained and displayed.The traditional image resizing methods include scaling with the interpolation al-gorithm and cropping. Since the sizes of the input image and the display device are different, the scaling method may deform the region of interest, and thereby leading to bad results. On the other hand, cropping may take the risk of removing the important content of the original image. It also pays the cost of losing much information of the original image. Recently, in order to overcome the drawback of the traditional methods, a variety of content aware image resizing methods are developed, which can adaptively recognize and preserve the important content after resizing. The content aware image resizing methods mostly consist of two main steps:firstly, several feature information, such as gradient, saliency, is captured from the original digital image, and an impor-tance map is defined; then according to the importance map, the pixels are removed or added with interpolation algorithm until the target size is reached.Based on the state-of-the-art literatures about image resizing, in this paper we present a novel intelligent image resizing method which is based on the visual attention mechanism. The main work of this thesis is summarized as follows:(1) A variety of traditional and state-of-the-art image resizing algorithms are inves-tigated and discussed in depth. In addition, their differences, advantages and disadvantages are also compared with the standard data sets of image resizing.(2) A novel boundary graph and saliency based adaptive image resizing method is proposed, and tested on the standard data sets of image resizing. In addition, we also define a novel quantitative index, which can be used to evaluate the resiz-ing results. The proposed method can well preserve the important content that people are concerned with in the original image during the process of resizing. Furthermore, the proposed method is compared with several state-of-the-art de-striping methods. The experimental results and quantitative analysis demonstrate its effectiveness and superiority.