"Color Distributions/Descriptors" is a topic in computer vision which is discussed in CVonline [1]

Color descriptors are low level descriptions of the color distribution in an image. Color descriptors, in addition to texture, shape, motion and location descriptors, are known as visual descriptors, which provide information about an image's contents.

Seven types of color descriptors are defined in the MPEG-7 standard.

Overview

edit

The MPEG-7 standard deals with describing visual and audio data. The metadata can be used for fast and efficient searching of these data formats.

The color descriptors represent color distributions and the relations between the colors in groups of images.

Color Space

edit

The color space descriptor defines what color model the image is using.

The MPEG-7 standard supports the following color spaces:

The linear transformation matrix color space allows a custom color space to be defined as a transformation of the RGB color space.

Color Quantization

edit

Color quantization is the process of limiting an image's number of unique colors, producing a palette of colors from which the image can be reproduced.

The color quantization descriptor defines how many unique colors the image has been reduced to, as a result of color quantization. This is used to produce a number of discrete color bins. The colors in each pixel are then assigned to the color bin closest to the original color. This descriptor is only used with the Dominant Color Descriptor. The other descriptors quantize the image into fixed amounts of color bins[3].


Dominant Colors

edit

This is a description of the dominant colors in the image. It includes the number of dominant colors, the color components of each dominant color, and the percentage of pixels in the image that correspond to each color.

This descriptor is useful when searching for images based on color values.

Scalable Color

edit

The scalable color descriptor is a histogram of the colors in the image, encoded by a Haar transform.

The histogram is generated by color quantizing the image into 256 bins in the HSV color space, with 16 bins for hue, and 4 bins each for saturation and value.

The image on the right is an example of color images with corresponding histograms. The left image and center image are more similar to each other based on the color histograms, compared to the image on the right.

This descriptor is useful when searching for similarities between images.

Color Layout

edit
 
Zigzag scanning

The color layout descriptor describes the spatial distribution of color in the image. The result is a compact descriptor which can be used in high-speed image matching and retrieval.

To generate the CLD, the image is divided into an 8x8 grid. An 8x8 pixel image is created, with each pixel given the representative color from the corresponding grid area in the image. The 8x8 matrix is transformed with the discrete cosine transform.

Finally, a zigzag scan is performed on the matrix. The resulting matrices (one for each color component) make up the descriptor.

The CLD has several advantages over the previous descriptors:

  • It can be used for matching images with little color data, such as handwriting or sketches.
  • It is scale/resolution invariant since all images, regardless of size, are divided into the same 8x8 grid.
  • The descriptor is only 8 bytes in size, so the required resources for calculations are very low. This means it can be used on devices with limited resources, such as mobile phones.
  • Due to its speed and accuracy, it is also used for matching videos.


Color Structure

edit

The color structure descriptor describes color features, including both the color of features and the structure of features. Unlike the scalable color descriptor, this descriptor is able to distinguish between two images with identical color histograms, but with colors in different areas of the image.

This is done by considering areas of 8x8 pixels and their colors, rather than individual pixels. By moving the 8x8 structure around the image, the presence of a color in the structure increases the color bin count in the histogram. It is the presence of the color which increases the count, not the number of pixels with that color.

This descriptor is always represented in the HMMD color space[4].

Group of Frames/Group of Pictures

edit

This descriptor is similar to the scalable color descriptor, except that the color histogram describes a video segment or a group of images.

The descriptor includes 2 bits to define how the histograms of the individual frames/images were combined to generate the final GoF/GoP descriptor. The three possible values are:

  • Average - the counts of each color bin in all the images are averaged
  • Median - using the median each color bin's count, the effect of outliers can be reduced
  • Intersection - the lowest count of each color bin is used. This method finds the minimum common colors in the group of images. Only pixels with the same location and color that are present in all the images will be counted in the histogram.

This descriptor is especially useful when matching images to videos, or images to groups of images.

References

edit
  1. ^ R. B. Fisher, "CVonline: an overview", Int. Assoc. of Pat. Recog. Newsletter, 27(2), April 2005.
  2. ^ Hyeon Jun Kim, "Method for quantizing colors using hue, min, max, difference (HMMD) color space", European Patent Office, November 1999.
  3. ^ Roshini Johr, "Color Descriptors from Compressed Images" [1]
  4. ^ Jens-Rainer Ohm, "The MPEG-7 Color Descriptors", Mitsubishi Electric R&D Centre Europe - France [2]

Bibliography

edit
  • Johr, Roshini. "Color Descriptors from Compressed Images" (PDF).
  • Cieplinski, Leszek (2001). "MPEG-7 Color Descriptors and Their Applications". CAIP '01 Proceedings of the 9th International Conference on Computer Analysis of Images and Patterns. Lecture Notes in Computer Science. 2124. Springer-Verlag GmbH: 11–20. doi:10.1007/3-540-44692-3_3. ISBN 978-3-540-42513-7.


edit

Category:Image processing Category:Artificial intelligence