Study On The Optimization Of Image Content And Structure

The optimization of image content and construction include two aspects:one is to optimize the distribution of color saturation or brightness in the image and make more visual information that can be perceived by human visual system, the other one is optimize the spatial distribution of visual features in the image and make the composition of the image be more pleasant. As an important issue in computer vision, it had aroused a wide range of discussions and studies in recent years. In this paper, we focus on the two child problems in this field which are image definition optimization and image visual attention learning.Image definition optimization is one type of image content optimization, its core content is image deblur. In this paper, firstly we partition the blur image into blocks and learn each block’s deblur degree and entropy information, secondly according to their relative relationship to compute each deblur parameters, through anisotropic and shock filter to estimate the PSF and get the latent block by using the reverse convolution, finally guided image filter is used to get the deblur image. Experimental results show that our method is adaptable for different types of blur images and outperforms than traditional methods.Image visual attention learning is very important in image automatic cropping. In this paper, based on the opponent and feedback mechanism in human visual system, we extract different opponent features in images, and both low-level and high-level information are considered into our saliency model to achieve a better prediction results.Experimental results show that the proposed methods are more adaptable and efficiency than traditional methods,and also optimize the image’s content and construction better.