Studies On Saliency-Based Image Enhancement

Digital image is a common digital information media.In recent years,with the widespread of smartphones and videos surveillances,the number of digital images daily produced is increasing gradually.However,due to improper exposure during the capture and the dynamic range limitation of sensors,the images produced are of poor quality,which makes it tough for digital image based applications such as automated driving and criminal investigation.Image enhancement methods are developed to solve the aforementioned problems.There are various of image enhancement methods were proposed these years.The paper summed up the problem that existing algorithms have in recent years by analysing existing digital image enhancement methods theoretically and via experiments: existing methods can hardly deal with images with lowlight and non-uniform illuminant input images,and are prone to loss naturalness during enhancement.Besides,these methods perform poorly while enhancing images taken under scenes of high dynamic range(HDR);Moreover,existing digital image enhancement methods rarely take visual saliency into account,resulting in visually unpleasing result.Taken the above issues into account,the thesis propose a saliency-based image enhancement method to solve the problems mentioned above effectively.The method in the paper utilizes a saliency-based method to fuse image exposure correction results,followed by a saliencybased method to optimize the image.Finally,a dynamic range extension algorithm was applied to generate the final output.The main work and contributions from this thesis are:(1)After analysing many related researches and experimenting on on-going image enhancement,the thesis proposed a saliency-based exposure fusion and image contrast enhancement method,and a dynamic range expansion algorithm framework which is based on HDR image generation and tone-mapping was also proposed;(2)Secondly,we proposed a degraded image dataset,and experimented with existing image enhancement methods on common datasets and the dataset proposed herein using subjective test and related metrics testing.The experimental results show that the method proposed in the thesis performs well on degraded images such as lowlight images as well as non-uniform illuminant images;(3)Finally,the thesis proved that the proposed saliency-based method is able to enhance input images of non-uniform luminance effectively via theoretical analysis and many comparison experiments.Moreover,the proposed algorithm enjoys a short execute time,and has advantage over most existing methods compared.