Research On Image Aesthetic Quality Assessment And Adaptive Enhancement

With the rapid development and popularization of digital image capture devices,particularly mobile devices such as smart phones and pads,people pay more attention to the aesthetic quality of photographic images.As an important and challenging topic in computer vision,computational image aesthetics aims to assess the aesthetics of images automatically as human being dose.In this paper,we focus on the intelligent analysis and adaptive enhancement of image aesthetic quality.We get a series results on aesthetic quality assessment and enhancement of variety image composition and visual attributes.The main research work is followed:(1)In the image composition quality assessment part,we give the formula for the depth of field in an image for the first time.On the basis of the existing visual balance evaluation algorithm,we add an evaluation item depicting the proportion of the salient area in an image,which effectively improves the accuracy.(2)In the visual attributes quality assessment part,we propose an adaptive window based clarity evaluation method,which performs more accurate evaluation for low depth of field images as well as normal images.Furthermore,a method for contrast full-scale evaluation using histogram analysis is proposed.With the observation of a correlation between color intensity and saturation,we use logistic regression to estimate the optimal saturation of an image,and then a saturation rating is computed based on the difference between the optimal and real saturations.(3)In the contrast enhancement part,we propose the resist over-enhancement contrast limited adaptive bi-histogram equalization(ROCLABHE).The adaptive division based on contrast is applied to resist local contrast over-enhancement and accelerates the calculation speed.In order to maintain the brightness of the images,we use minimum mean brightness error bi-histogram equalization(MMBEBHE)to enhance the local areas.Furthermore,a comprehensive contrast enhancement algorithm is proposed,which selects the enhancement algorithm adaptively according to the contrast evaluation results.The algorithm forms a complete system of image contrast evaluation and adaptive enhancement.(4)In the clarity enhancement part,we propose a new image sharpening algorithm based on wavelet transform.The low frequency is sharpened based on the edge extracted in multiscale and the high frequency is denoised.Then the image clarity is improved and the noise amplification is suppressed doubly.(5)In the comprehensive experiment part,we use SVM classifier for the image classification experiments in the image database,and adopt weighted sum combined with threshold segmentation for the comprehensive evaluation of the aesthetic quality from a new point of view.A higher classification accuracy is obtained.We get the exact score of the aesthetic quality and the specific results of all aspects,so as to achieve the aesthetic analysis of the image.Finally,we enhance the low aesthetic images and evaluate them using the proposed algorithms.The experiment results show that the proposed algorithms are effective.