Visual Adaptive Image Encoding Based On Just Noticable Distortion

In the imformation age nowadays, the exchanging of imformation is getting more frequent, widespread, and easy. Visual imformation is an important and the most convenient way of getting imformation, for the imformation contained in visual contents(picture) is aboudent in quantity, which would require a lot of time and space to represent in the way of language or text. To express an image in digital manner effectively and acuritly, requires not only certain loss-less entropy encoding measures to remove the imformation redundency in the picture, but also certain lossy measures to achieve better efficiency without visual quality degrading. And to investigate the property of human visual system, obtain an model that unveils the tolerable noise that could be hidden in a varient of images, is an key issue in visual adaptive image encoding research. The tolerable noise is called Just-Noticable-Distortion, or Just-Noticable-Difference, or JND for short. A proper JND model is used to predict the JND profile of a certain image.In the first part of this thesis, the development of image compressing is briefly introduced, pointing out that the application of visual adaptive image encoding is nessesary, and the research status of JND modeling is also introduced.The lumination adaptive aspect of image domain JND model is explained later, by illustrating the pysiology of it. A well known lumination adaptive model is discussed, and improved by parameterizing the model, making it adjustable, and become suitable for varias of experiment environments with different parameters.The Key part of this thesis is on the texture masking effect, and edge masking effect, what consistes the contrast masking effect. The edge masking phenominant is explained. Inspired by the feature integration hypothesis and cognition knowledge, a hypothesis is proposed, that tries to explain why the noise is more sensable about the edge region and less sensable in the texture region.Based on the hypothesis, an innovative edge detecting method simulating the primary visual cognition process is proposed, and later integrated into JND model. It is discovered in subjective quality measurements that the hypothesis is reasonable, and such method used in JND model is effective.In the last part, JPEG image compressing standard is introduced, and a scheme of integrating JND prediction model into JPEG is proposed, gives an example of visual adaptive image encoding method. The experiment result is discussed.Critics on methods in JND study is also included.