Graphics Reference
In-Depth Information
SIFT is based scale selection of image features, and it establishes multi-scale space,
and detects same feature point at different scales, then determines the location of the
feature point in their scales simultaneously determined to achieve the scale of anti-
scaling purposes, bound by a number of points as well as low-contrast edge response
points and rotation invariant feature descriptor extraction in order to achieve the pur-
pose of anti-affine transformation. The algorithm mainly includes four steps:
(1) To establish scale space, and to look for candidate points;
(2) To confirm key points accurately, and to eliminate unstable points;
(3) To acquire direction of key points;
(4) To extract feature descriptors.
Results of SIFT feature extraction are shown in Figure 1.
Fig. 1. Match Points of SIFT Features
Since SIFT has a good unique and rich amount of information for the reference im-
age and degraded image feature matching between, it is particularly suitable for image
quality assessment characteristic parameters. At the same time it has a scalable, can
be very convenient feature vectors with other forms of joint, so consider using SIFT
and structural similarity with the method to compensate for image features SSIM
algorithm does not consider defects.
2.2
Framework of Improved Algorithm
The paper fusions SIFT features into comparison of luminance, contrast and structure,
and a new metric of FR-IQA is presented. Diagram of the SIF-SSIM measurement
system is shown in Figure 2.
Search WWH ::




Custom Search