Using Gray Level Histogram Specification to Improve the Quality of Digital Image
Ning Yu

ningyu@siu.edu

Gray Level Image

Usually, the images are captured as a color image, not a gray level image by common digitizing devices, such as Digital Camera. In term of the property of light intensity, the color image is converted to the gray level image according to the following equation. G = R*0.299+G*0.587+B*0.144

The following graph is the color image.

Figure 1. The Original Image.

Figure 2 is the image converted from the color image. For true color images, there are 24bits for every pixel. Red, green and blue separately occupy one byte. Gray level images are received by calculating the R, G and B. And the gray level value is the result of the calculation of priority RGB at every pixel.

Figure 2. The Gray Level Image.

Figure 3 is the histogram of the gray level. The histogram indicates that most of the pixels are too dark. Only the minority of pixels are light. So what we need to do is to make the image clear to view the scene and the character.

Figure 3. The Histogram of Figure1

Histogram Equalization

The histogram-equalized image will span a fuller range of the gray scale without the need for further parameter specifications. The Figure 4 is the result of histogram equalization.

Figure 4. The Image Through Histogram Equalization.

And the histogram graph is in Figure 5. In figure 4, we can discern the scene, but the character’s appearance can not be distinguished. Our goal is to discern both of the foreground and background. So, although histogram equalization can scale the histogram linearly and smoothly and do it automatically, the result of image processing does not satisfy our need.

Figure 5. The Histogram of Figure 4.

Histogram Matching

Histogram Matching, also called Histogram Specification. As indicated in the preceding discussion, histogram equalization can determine a transformation function automatically. Although automatic enhancement is desired and it is a good approach, in many situation, E.g. the forementioned figure, the histogram equalization is not the best approach. In particular, it is useful sometimes to be able to specify the shape of the histogram that we could have the best processed image. The method in which we specify the histogram is called histogram matching.

The Figure 6 is manually specified.

Figure 6. The Specified Transform Function.

The Figure 7 is the result of histogram specification.

Figure 7. The Image Through Histogram Specification.

The Figure 8 is the histogram of histogram specification.

Figure 8. The Histogram of Figure 7.

Noise Analysis

Obviously, some noises exist in the image. How to eliminate this noise pollution should be considered. The following graph is the result of using Median Filter.

Figure 9. The graph of applying the Median filter.

The noises are still there. Analyzing the original image, we know this is a picture taken at night, which is similar to the pictures taken by the astronomical telescope in the dark space. So we can guess that noises are superimposed by the camera sensor. To solve the problem of sensor noise, the best way is to average lots of images or increase the time of exposure of camera sensor. To illustrate the technology of the image enhancement, I think, the effect is enough.

Conclusion

We can draw a conclusion that histogram equalization is a good approach for digital image processing in spatial domain, having the histogram more smoothly and making the image clearer. However, for some specific images, we still need to specify the particular transformation functions to receive the best processing result. It depends on your need and the specific images. In sum, histogram approach is a good method to help professionals obtain the details of images in spatial domain, which is very significant for researching and studying.