Automatic Image Equalization for Texturing in 3D Models

The “texturing” procedure in 3D modelling is a common operation utilized to add color information to the geometric model and thus increasing the quality of the model by bringing it closet to the real object. This is done by assigning the color information to the different parts of the model using digital images of the object itself. A number of images is taken with a camera in order to be able to cover the model completely. Such procedure can be described appropriately as extracting parts of images and associating them to a spatial coordinate which corresponds to a coordinate of the model. There are a series of geometric and chromatic difficulties to carry out this task; the former consist of adapting 2D images to a 3D model, the latter are caused by different illumination and camera settings. This last problem brings forward a question of color consistency between adjacent images. Chromatic inconsistency between neighbouring images in the model is evident during the visualization as there is a very clear-cut border line which lowers the quality of the model. This paper presents a method for equalizing the colors between neighbouring images. The method proposed in this paper consists of a procedure which analyses neighbouring image pairs to evaluate color inconsistency and automatically apply an algorithm for equalization. This decreases significantly the color discrepancy along the border between image pairs. The model which is used is in Web3D (VRML) format, and it is processed in order to extract control points between the neighbouring image pairs. This produces a matrix with color values of equivalent1 pixels between pairs. Utilizing correlation analysis and grey value frequency distribution a Look Up Table is calculated for transformation of one image (called “slave”) and the other image taken as reference image (called “master”). Interesting results have been obtained especially considering the perception of the model when it is visualized by an operator. Matter of factly the color difference is not seen anymore when the model is projected on the screen. The quality of the representation is therefore improved and definitively more realistic. This post-processing procedure for image color equalization can be considered a valid improvement in the overall modelling procedure.