For my academic project this year that involves some image processing requirements, I am given a large number of images (about 86,000) images, each of 5 Megapixels (2480x2480), in which I need to separate out black and non-black images.
My team's proposed approach is that we scale down the 2480x2480 images to 240x240 (by picking an arbitrary pixel every, lets say 8 pixels) and then plotting a histogram of pixel value vs no.of pixels, where the pixel value of 0 is black and the highest pixel value (e.g. 255 for 8 bits per pixel) is white. If the mean of the histogram is very close to 0 and if the variance from this value of the mean is very small (to make sure we don't include images of low brightness in the black category), we categorize the image as a black image.
My team is wondering, do you know of a fastest image processing approach to identify if an image is purely black or is our above approach fine (please comment if theres any mistakes in our approach)?
Dear Mohammed Ibrahim ,
It seems to me that your team's approach is fine. But I think there is no need to scale down the images to 240x240. Additionally you have to plan for range of mean (0 to what extent for black) and upper limit of variance.
By the way which programming language you are going to use?
( Please note that I have little experience in MATLAB image processing. )
Dear Mohammed
You can find black pixel easily.
Plz use following code in matlab
%==========================
f=find(I>=0);
f2=find(I(f)
Try SIFT algorithm or use in conjunction with SVD ! What you are trying is not an efficient method for huge data set.
@ Subrota Halder : we are using python
@ Mostafa: method is simple does sound inefficient in terms of speed but i will implement it
@ hrebesh : could you elaborate on this approach ?
- Badges
- Science topic
- Similar topics
- Computer Science
- Images