TOP  >  Thesis/Dissertation  >  Estimation of Potential Sunny and Shadow Regions by Analyzing their Spatial Continuity and Periodicity

Color Correction for Person Re-identification in Nonuniform Lighting Condition

In this thesis, we propose a method that corrects nonuniform color variation by direct light for the Person Re-identification problem. Since color features change by illumination condition and optical camera settings, correspondence of color distribution between two cameras must be acquired. To realize this, some methods correct color distribution to the one under standard illumination for each camera in advance. However, these methods assume that color variation occurs uniformly over an image. We propose the method that corrects color variation in sunny and shade conditions in the scene. To divide these condition about both person and background on images, we analyze observed images over a certain long term. For a person image, based on its position and color, we select one image adequate for correction. For a background image, we select an image adequate for standard illumination and acquire sunny and shade regions by comparing regions under the much different illumination. Finally we acquire Brightness Transfer Function from the two background regions and apply it to the selected person image according to its condition. Experimental results show the accuracy of our color correction itself and the effectiveness of the correction for the Person Re-identification problem.