Correcting the Geometry and Color of Digital Images

This work is funded by NSF.


IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 19, no. 10, pp. 1152-1158, 1997.


Abstract:

A unified method for correction of color and geometry of digital images is presented. The method requires the use of a color chart which contains an array of known color patches. Knowing the true colors and geometry of the chart and the colors and geometry of an image of the chart, two transformation functions are determined to represent the color and geometry characteristics of the image acquisition system. The transformation functions are then used to correct the color and geometry of obtained images. Elastic parametric functions which adapt well to the local color and geometry characteristics of the camera are used to represent the transformation functions. Experimental results show that the proposed method effectively corrects the color and geometry of images obtained under different scene lightings and camera parameters.

Results:

This is the standard image.

(left) Original daylight image with f stop = -2. (middle) Image after color correction.
(right) Image after geometry correction.

(left) Daylight image with f stop = -1. (middle) Color corrected image.
(right) Geometry corrected image.

Daylight, f stop = 0.

Daylight, f stop = +1.

Daylight, f stop = +2.

Electronic Flash, f stop = -2.

Electronic Flash, f stop = -1.

Electronic Flash, f stop = 0.

Electronic Flash, f stop = +1.

Electronic Flash, f stop = +2.

Tungsten, f stop = -2.

Tungsten, f stop = -1.

Tungsten, f stop = 0.

Tungsten, f stop = +1.

Tungsten, f stop +2.

Fluorescent, f stop = -2.

Fluorescent, f stop = -1.

Fluorescent, f stop = 0.

Fluorescent, f stop = +1.

Fluorescent, f stop = +2.