Color-based Feature Detectors and Descriptors

Overview

This page is focused on the use of color in local feature detection and description.

Feature Detection

Laplacian-of-Gaussian Detector: From luminance to color

Salience refers to the physical, bottom-up  distinctiveness of an object. In computer vision, we use salience to focus on regions of interest and not the whole image,  thus reducing the computational complexity. Traditionally, state-of-the-art methods for feature detection exploit only the shape salience, however the color salience provides more discriminative regions. This is illustrated in the following figure.

(a) input image

(b) shape salience-based detection

(c) color salience-based detection

Detecting Salient Contours

The information obtained from image edges are very important for many tasks in computer vision, such as: object detection, segmentation and recognition. Luminance edges are influenced by shadows, shadings and highlights. Thus, color edges provide highly informative regions for image analysis and understanding. This is illustred in the following figures.

(a) input image	(b) color edges detected	(c) color-boosted edges detected
(a) input image	(b) color edges detected	(c) color-boosted edges detected

This file contains the Laplacian-of-Gaussian detector, Color-boosted Laplacian-of-Gaussian detector and the color-boosted edges detector. Download.

Format

points = ColorLoG(imname, alpha, thresh)

Inputs:

imname - image name

alpha - salience parameter, shape salience (alpha=0) and color salience (alpha=1)

thresh - threshold's detector (default 80)

Output:

points - coordinates and scales detected [num_regions][x y scale]

Example:

points=ColorLoG('0221.jpg',0);

Feature Description

Experiments have shown that a combination of color and shape descriptors outperforms a pure shape-based approach.

Hue Descriptor

Hue histogram descriptor is an photometric invariant color descriptor. Download.

Format

descriptors = hue_extraction(points, image, psize, nbins, sflag)

Inputs:

points - coordinates and scales detected [num_regions][x y scale]

image - color image

psize - patch size or neibourhood of features

nbits - number of bits of hue descriptor

sflag - amount of smoothing of final histogram

Output:

descriptors - hue descriptor list

Example:

hue_desc = hue_extraction(points, im, 20, 36, 2);

D. Rojas Vigo, F. Shahbaz Khan, J. van de Weijer, Th. Gevers, The Impact of Color on Bag-of-Words based Object Recognition, 20th International Conference on Pattern Recognition, Turkey, August 2010 (accepted).

D. Rojas Vigo, J. van de Weijer, Th. Gevers, Color Edge Saliency Boosting using Natural Image Statistics, IS&T's 5th European Conference on Colour in Graphics, Imaging, and Vision (CGIV), Finland, June 2010 (accepted).

J. van de Weijer, C. Schmid, Coloring Local Feature Extraction, Proc. ECCV, Part II, 334-348, Graz, Austria, 2006

J. van de Weijer, Th. Gevers and A.D. Bagdanov, Boosting Color Saliency in Image Feature Detection, IEEE Trans. Pattern Analysis and Machine Intelligence (PAMI) , vol. 28 (1): 150-156, January 2006