Dense Scale Invariant Feature Transform-Based Quality Assessment for Tone Mapping Image

Abstract

High dynamic range (HDR) images give the human visual system (HVS) a better visual experience due to their wide luminance range. However, traditional display devices can't capture such a large luminance dynamic range, which needs to be remedied by tone mapping (TM) operation. In addition, TM images usually lose a lot of detail information in low luminance and high luminance areas, So a TM image quality evaluation method based on dense scale invariant feature transform (DSIFT) is proposed in this work. First, the DFIFT descriptors of the HDR image and the TM image are extracted respectively, and the local similarity is calculated to represent the detail loss of the TM image. Then, the local quality map is refined by using the Gauss exposure curve because HVS is more sensitive to the detail loss in low-dark and highlight areas of the TM images. Finally, considering the color distortion characteristics of TM image, an objective perception quality evaluation model is established. Experimental results on a public database demonstrate that the proposed method is in good agreement with human visual perception.

Iranian kinect face database (IKFDB): a color-depth based face database collected by kinect v.2 sensor

“…This feature is fast and robust, and works well on Depth images. This method is edge based and similar to Scale-Invariant Feature Transform descriptors (SIFT) [52], and shape contexts, but differs is computed on a dense grid of uniformly spaced cells and uses overlapping local contrast normalization for improved accuracy [37].…”