Color Cast Detection Method Based on Multi-feature Extraction - Advances in Image and Graphics Technologies - page 103

Color Cast Detection Method
Based on Multi-feature Extraction

Minjing Miao 1 , Yuan Yuan 1 , Juhua Liu 1 , and Hanfei Yi 2

1 School of Printing and Packaging, Wuhan University
2 College of Physical Science and Technology, Huazhong Normal University
430079 Wuhan, China
[e-mail]

Abstract. In order to raise the accuracy rate of the color cast detection and to
make the method universal, the paper carries out a color cast detection method
based on multi-feature extraction. Firstly, calculate the four features that are the
textural property of the luminance channel, color numbers, histogram of RGB
color space and statistical characteristics of the Gabor filter, then use AdaBoost
to train and classify. The experiment will be done using 11346 images in the
Ciurea database. The result shows that this method has a low error rate and good
classification results, which is universal to natural images taken by cameras.

Keywords: Multi-feature extraction, AdaBoost;color cast detection.

1 Introduction

When capturing an image, the camera is easy to be influenced by the illumination,the
reflective properties of the object itself and the photosensitive coefficient of the image
capturing devices. Thus, the color of the obtained image is different from the real
color of the object, which is called the image color cast [1].And it could have a bad
affection on the human visual perception. So it is important to do image color cast
detection for assessing the image quality.
Nowadays the color cast detection methods are grouping into 2 parts, that are the
subjective judgment and the objective classify metrics. In practice, however, subjective
evaluation is usually too inconvenient, time-consuming and expensive. In the past dec-
ade, many objective color cast detection metrics have been carried out. In general, the
metrics can be classified into two groups: algorithms based on the deviation of chroma-
ticity information and algorithms that use mathematical statistics to classify the images.
Examples of the first group are the White-Patch algorithm [1], the Grey-World algo-
rithm [2], the dimensional histogram statistic algorithm [3] and the equivalent circle
algorithm [4]. Allabove color cast detection methods are based on specific imaging
assumptions. These assumptions include the set of possible light sources, the spatial and
spectral characteristics of scenes, or other presumptions (e.g. white patch, averaged
color is grey, etc.). As a consequence, no algorithm can be considered as universal.