[Home]

Camera Gamma Correction


In order to compare the outputs of color balancing algorithms with the camera outputs, we need to cancel out the camera functions except color balancing. In this project, we assume all other camera functions other than color balancing can be approximated as gamma function. To cancel out other functions of a camera, the output images of the camera must be passed through the inverse camera gamma correction before being compared with algorithm outputs. In addition, to do comparison in RGB space, the camera gamma correction has to be done for R, G & B separately. Thus the camera gamma functions need to be obtained for R, G & B respectively.

We take the following steps to evaluate camera functions:

1. Generate a GrayPatch image with nine gray levels that evenly distribute from 0 to 256.
2. Do inverse monitor gamma correction for GrayPatch image.
3. Take a picture of GrayPatch image shown on the monitor.
4. Approximate camera gamma functions for R, G & B.

We describe the results of the above steps in details:

1. Generate a GrayPatch image with nine gray levels that evenly distribute from 0 to 256 by GrayPatchFun.m.


2. Do inverse monitor gamma correction for GrayPatch image by GrayPatchFun.m.


3. Take a picture of GrayPatch image shown on the monitor.


4. Approximate camera gamma functions for R, G & B by CameraGamma.m.

Since the output image of the camera distorts in shape, so the target gray patch areas do not fill in the whole screen, we use nine test patches to locate the center of each gray patches and use the center area of them with size 210*300 as our test areas.

After getting the mean R, G & B values for each gray patch, we estimate camera gamma function by fmin function in Matlab. The equation of camera gamma we use is of the form:


We plot the actual camera output and the estimated gamma function for R, G & B values as following:



This is the result of the camera evaluation:

R G B
a
0.8694
0.8990
0.9215
gamma
1.4870
1.6327
1.6565

The above estimations are not an exact match, but close enough. We can use them to do inverse camera gamma correction for R, G & B respectively.


trek@alumni.stanford.edu
lihui@leland.stanford.edu