Feature Combination beyond Basic Arithmetics

Abstract

Kernel-based feature combination techniques such as Multiple Kernel Learning use arithmetical operations to linearly combine different kernels. We have observed that the kernel distributions of different features are usually very different. We argue that the similarity distributions amongst the data points for a given dataset should not change with their representation features and propose the concept of relative kernel distribution invariance (RKDI). We have developed a very simple histogram matching based technique to achieve RKDI by transforming the kernels to a canonical distribution. We have performed extensive experiments on various computer vision and machine learning datasets and show that calibrating the kernels to an empirically chosen canonical space before they are combined can always achieve a performance gain over state-of-art methods. As histogram matching is a remarkably simple and robust technique, the new method is universally applicable to kernel-based feature combination.

GlandVision: A Novel Polar Space Random Field Model for Glandular Biological Structure Detection

“…However, when we combine these two methods by simply summing their outputs together, our result exceeds part-based model. Besides this, if we adopt the method introduced in [12] to combine the two outputs, our result is further improved. We directly use the histogram matching code by [12] which is publicly available.…”

“…We directly sum the two outputs together. Recently, Fu et al [12] have shown that calibrating the distributions of different measures before combining them can improve performances and we have also tried this approach in the experimental section. The flowchart of our complete gland detection algorithm is depicted in Fig.5.…”

Random Forest for Image Annotation

“…It has been a common practice [5,4] to use multiple kinds of features to represent one image. How to efficiently combine different features is a non-trivial task [20]. Existing methods include using an equal weight for different features [5], a distance specific weight for each feature [4], etc.…”