A probabilistic definition of salient regions for image matching

Abstract

Abstract. A probabilistic definition of saliency is given in a form suitable for applications to image matching. In order to make this definition, the values of the pixels in pairs of matching regions are modeled using an elliptically symmetric distribution (ESD). The values of the pixels in background pairs of regions are also modeled using an ESD. If a region is given in one image, then the conditional probability density function for the pixel values in a matching region can be calculated. The saliency of the given region is defined to be the Kullback-Leibler divergence between this conditional pdf and a background conditional pdf. Experiments carried out using images in the Middlebury stereo database show that if the salience of a given image region is high, then there are relatively few background regions that have a better match to the given region than the true matching region.

Saliency propagation from simple to difficult

“…For example, Maybank [19] proposed a probabilistic definition of salient image regions for image matching. Yang et al [28] combined dictionary learning and Conditional Random Fields (CRFs) to generate discriminative representation of target-specific objects.…”

Accuracy Analysis and Structure Configuration of Binocular Vision System

“…Generally, the system uses human experience [1,2] to be built. At the same time, the current research of binocular vision system is focused on camera calibration [3], image matching [4] and other important aspects, lacking of systematic measurement accuracy analysis. Thus, the study of the structure of binocular vision system configuration and precision analysis is important, helping to improve the measurement accuracy and its application.…”

Stereo matching-based definition of saliency via sample-based Kullback–Leibler divergence estimation

“…They aim to locate predefined attributes, such as corners and boundaries [5,6], local extrema [7,8], or complex image regions [9][10][11][12]. The topdown methods are closely linked to particular applications such as object recognition [13][14][15], object detection [16], and image matching [17]. The top-down definition of saliency is used to measure the effectiveness of a feature for a particular task, i.e., the high saliency of a feature in [13] indicates the feature is discriminative for the given object category from the other categories, and the high saliency of a region in [17] means the region has a high probability to be correctly matched.…”

“…The saliency of a region is the Kullback-Leibler (K-L) divergence from the conditional pdf for the matching region to a background pdf. Our goal is to reduce the complexity of the calculations in [17]. To this end, we directly model the conditional pdfs without an intermediate estimation of the joint pdfs.…”

“…Without losing accuracy, the conditional pdfs are directly estimated using the samples and avoiding the calculation of joint pdfs [17]. The K-L divergence between two pdfs is also directly estimated from the samples.…”