Skip to main content
Figure 3 | EURASIP Journal on Image and Video Processing

Figure 3

From: An efficient approach for robust multimodal retinal image registration based on UR-SIFT features and PIIFD descriptors

Figure 3

Non-uniform distribution of features. Original SIFT point detector in different scale layers of octaves for color and corresponding FA images for T c = 0.03. It is required to set T c manually for various retinal image pairs to achieve repeatable points. (a, b) Image-space distribution. The original SIFT fails to extract adequate features or distinctive ones in color images while detects a lot of redundant features in FA images (boundary points are masked). (c, d) Scale-space distribution. There are inadequate features in some scales which decrease the scale invariance image matching.

Back to article page