dresa
commented
6 years ago Hi Hossein,
Happy to hear you're interested. Indeed, it's possible to improve the prediction accuracy for photo-quality.
In your first image (blue-purple), I think I see a segmentation method that is able to distinguish between objects and background. If so, I think many image-quality measures could use that type of segmentation information. Ignoring background and concentrating on foreground objects is something I have been planning to do. What is the filter/method you have used? What I know from scientific literature, this type of segmentation methodology could be used as a backbone to many quality-measures.
The image segmentation method that I have used is based on clustering pixels in CIELUV color space, which makes the segments somewhat scattered. In your second image I see an effective edge-detection filter. It could be used to develop another image segmentation algorithm, which better recognizes coherent areas as segments, and detects meaningful objects. It would be helpful to have the edge-detection function available in the derivedimage.R file, accepting as input an RGB image as m x n x 3 array, and returning an m x n matrix with edge-intensity values. The next step would be finding a suitable segmentation method in the literature.
Thanks for the images -- I see potential in your methodology. If possible, I'd like to see these methods added to the collection of image-quality methods. I think it's easiest to start with adding the edge-detection method to derivedimage.R, and perhaps later expanding to new segmentation methods, which could replace the current segmentation methodology and improve photo-quality predictions.
Kind regards, Esa
Dear Junttila, can I know that images which uploaded, can be useful for growing up your job? how?