Evaluation of Intensity and Color Corner Detectors for Affine Invariant Salient Regions

Abstract

Global features are commonly used to describe the image content. The problem with this approach is that these features cannot capture all parts of the image having different characteristics. Therefore, local computation of image information is necessary. By using salient points to represent local information, more discriminative features can be computed. This research is based on an existing affine invariant local feature detector, in which the features are assumed to be intensity corners. First, the existing algorithm is extended with the intensity based SUSAN corner detector which fundamentally differs from the original Harris corner detector. Second, the algorithm is extended to incorporate color information into the detection process. This results in a comparison between three different detection algorithms: the intensity based algorithm using the Harris or SUSAN detector and a color based algorithm that uses two color extended Harris detectors. The different algorithms are compared in terms of invariance and distinctiveness of the regions and computational complexity.

Adaptive Corner Detection Algorithm and its Extension using Window-based Approach for Gray-scale Images

“…Coleman, Scotney and Kerr [27] adopted an existing edge detection method and extended this work to corner detection using the Linear Gaussian product operator. Sebe et al [8] compared four different algorithms in terms of invariance and distinctiveness of the extracted regions and computational complexity; and proposed a color-based corner detection algorithm.…”

“…On the other hand, the intensity-based methods estimate a measure to detect corners directly from the Gray values of the original images without a prior segmentation. Rutkowski et al [1], Heyden and Rohr [2], Zheng et al [3], Schmid et al [4], Rockett [5], Tissainayagam and Suter [6], Mokhtarian and Mohanna [7], Sebe et al [8], and Dutta et al [9] provided good literature survey on the existing corner detection algorithms and evaluated their performances.…”

Heterogeneous sensor fusion framework for autonomous mobile robot obstacle avoidance

“…plain or blurred image regions. To this end, we build on the work in [25], [26] and employ the "Cornerness" of pixels in the image as a measurement of their high-frequency information content. Recall that the "Cornerness", as presented in [25], for a pixel with coordinates u on the image lattice is defined as…”

Sparse Color Interest Points for Image Retrieval and Object Categorization

“…These methods ignore saliency information contained in the color channels. However, it was shown that the distinctiveness of color-based interest points is larger, and therefore color is important when matching images [5]. Furthermore, color plays an important role in the pre-attentive stage in which features are detected [6], [7] as it is one of the elementary stimulus features [8].…”