Image lighting can change the perception of image characteristics which means that relative to image representations, light changes can result in huge image differences. Using a simple or conventional technique may fail, so how can we process image subjected to extensive light?
Hello, I agree with your opinion that light changes can results in huge image differences. For image process for example edge detection, a image should contain relatively obvious grayscale differences. However, for image captured under bright light, the grayscale is almost the same which is not ideal for image process. Therefore, certain image enhancement technologies should be used ahead of image process. However, not all details can be reconstructed so some robust image process algorithms should be used. Thank you!
Err, not really work all the time, but, isn't before we process the image, it is recommended to "standardize" the image first, like "widening" and "standardize" total colour value? Not working for photo's which too bright or dark though. :)
Some image processing algorithms are very sensitive to image brightness, while others are less sensitive. My suggestion is to use algorithms that are less sensitive to brightness. For example, we can use the Hue component of an image.
HSL or HSD image spaces are good to process images with extensive light. so convert RGB images to HSL or HSD images and separate out lightness/brightness and then normalize the lightness/brightness.
In too bright/dark images the histogram is skewed to one side. There are some ways to equalize the histogram and higher the quality. If you are using MATLAB for image processing you can use commands like "imadjust" or "histeq" to equalize the histogram in each dimension (in RGB images you have three dimensions for example 640x480x3 that you should work on third dimension here). Merging the equalized layer again will give you better quality image.
Histogram Processing is best choice in this case, therefore explore Histogram Equalization and Histogram Specification
In my opinion a solution can be equalize the image histogram, another can be normalize the image. You also can process the image in the frequency domain. Try to enhance the image adjusting the enhance process.
Depending on your goals, I'd recommend trying to not adjust the brightness (unless just for aesthetics) through equilization until you're sure that you can't extract the information you want. Here's an example of local adaptive thresholding that is fairly insensitive to contrast gradients:
http://scikit-image.org/docs/dev/auto_examples/plot_threshold_adaptive.html
Generally, histogram equilizaiton will prove a detriment to adaptive thresholding. Depending on your goals, this may be a concern.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing