Faculty of Informatics Vienna University of Technology Institute of Computer Aided Automation PRIP Home PRIP Home
Personal tools
You are here: Home People Julian Stöttinger Color Interest Points

Color Interest Points

Download scale invariant interest points based on arbritrary color spaces and PCA scale selection. Download binaries for linux 64 bit for detector in RGB, OCS, color boosted OCS, and HSI

Scale Invariant Color Corner Detector

Quickstart:

  1. Download the software here.
  2. Read the readme for usage.
  3. Regard the paper for details.

More sparse and distinct local features for object recognition and classification

 Local image descriptors computed in areas around salient points in images are essential for many algorithms in computer vision. Recent work suggests using as many salient points as possible. While sophisticated classifiers have been proposed to cope with the resulting large number of descriptors, processing this large amount of data is computationally costly.

We propose computational methods to compute salient points designed to allow a reduction in the number of salient points while maintaining state of the art performance in image retrieval and object recognition applications.

To obtain a more sparse description, a color salient point and scale determination framework is proposed operating on color spaces that have useful perceptual and saliency properties. This allows for the necessary discriminative points to be located, allowing a significant reduction in the number of salient points and obtaining an invariant (repeatability) and discriminative (distinctiveness) image description.

Example: VOC PASCAL 2007 Challenge Pic Nr. 337

VOC2007 Picture 337

Pic Nr. 337 - person and bike

VOC2007 Picture 337 Harris Laplacian

Harris Laplacian download software

VOC2007 Picture 337 Harris Hessian

Hessian Laplacian download software
 

VOC2007 Picture 337 ScIvHarris OCS

Scale Invariant Color Harris with PCA scale selection using color boosted OCS download software

VOC2007 Picture 337 ScIvHarris HSI

Scale Invariant Color Harris with PCA scale selection using HSI download software

Results

Scale invariant color interest points outperform state of the art detectors in various applications. Detailed results can be found in

Lonely but Attractive: Sparse Color Salient Points for Object Retrieval and Categorization
Julian Stöttinger, Allan Hanbury, Theo Gevers and Nicu Sebe, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), Workshop on Feature Detectors and Descriptors: The State Of The Art and Beyond, Miami, Florida, USA, June 20, 2009, pp 1-8.
Download article and citation:
[pdf] [bibtex]

Repeatability Results

On average, we gain stability in various challenges from the test sets from Krystian Mikolajczyk:

Repeatability on color color test sets

For lighting changes, the performance increases significantly:

rep_light.pngrep_light.png

Image Retrieval Results

The ALOI database provides 1000 classes of images of artefacts und various changes of light and position. Scale invariant color points oiutperform state of the art using significantly less interest points. Under change of illumination direction:

IllumDir_ROC_CVPR.png

Object rotation:

Object Rotation ALOI color interest points

Change of color temperature does not change too much of the pictures - greyscale based approaches are as stable as color interest points:

Coltemp_ROC.png

Comparison of number of points per approach under changing illumination direction:

ALOI Image Retrieval number of features

VOC 2007 object categorization

With fewer points, color interest points maintain state of the art performance:

VOC 2007 results color interest points

For more detailed results, please refer to my list of publications.

Download Software for Scale Invariant Color Harris

The software for linux 64bit can be downloaded here. It is very experimental - but it works for me.

Usage:

$./scivharris64

for help and standard parameters. Binary just reads jpgs!

Example

$./scivharris64 -i nicepic.jpg -o niceregions.out
just do it...

$./scivharris64 -i nicepic.jpg -o cleverregions.out -m 1200 -c HSI -s 1 -f 1.5 -n 12

extract regions in HSI color space with a maximum of 1200 extracted regions / corners. First Sigma is 1, scale factor is 1.5, and there are 12 scale steps.

Output file format is from Krystian Mikolajzyk. Refer to his great site for details.

For your convinience, I added some very simple bash scripts to process directories full of images. Combine it with Krystian's binaries and have fun!