Just Noticeable Difference for Images With Decomposition Model for Separating Edge and Textured Regions

Perceptual based SAO rate-distortion optimization method with a simplified JND model for H.265/HEVC

“…For the JND models reported by Yang et al [10] and Liu et al [11] in the pixel domain, four directional high-pass filters were used for texture detection and the Canny detector followed by a Gaussian low pass filter were used for edge detection. Although it has been demonstrated that Canny detector has better performance than that of Sobel detector in terms of signal to noise ratio and edge detection, especially under noise conditions [16], its algorithm implementation requires a series of steps including non-maximum suppression, Gaussian filter and so on, which incurs a significantly higher computational cost.…”

“…Although it has been demonstrated that Canny detector has better performance than that of Sobel detector in terms of signal to noise ratio and edge detection, especially under noise conditions [16], its algorithm implementation requires a series of steps including non-maximum suppression, Gaussian filter and so on, which incurs a significantly higher computational cost. Furthermore, the more recent JND model in [11] adds image decomposition to decompose an image into structural image and textural image to make the texture, masking component more accurate and improving the performance of JND performance in pixel domain significantly. However, this additional decompose process further increases the model complexity.…”

“…A JND model in the pixel domain usually takes account of luminance adaptation P L ½n and texture masking effect P T ½n and can be expressed as [7][8][9][10][11] JND p ½n ¼ P L ½nþP T ½nÀκ minðP L ½n; P T ½nÞ ð8Þ…”

“…Subband JND models have been investigated extensively among all JND functions with relatively fewer reports on JND models in the pixel domain. After Chou and Li first introduced the JND threshold model in [9], Yang et al [10] and Liu et al [11] reported their improved JND models in the pixel domain to improve model accuracy and to further remove the visual redundancy. However, a more accurate JND model is usually accompanied by an increase in computation complexity, representing more complex human visual characteristics.…”

Interpreting MOS scores, when can users see a difference? Understanding user experience differences for photo quality

“…For the subband-domain models, the image needs to be first transformed into a subband domain, such as the DCT-domain [9,21]. On the other hand, pixel-domain methods [13,19,24,26,29] are directly calculated on the spatial domain and thus are more convenient and less computationally complex. As a result, they have been used for a wide variety of applications such as visual quality assessment and enhancement [26].…”