I need a measure which results in 0 or 100 if 2 RGB images being tested are same, and give different value if there is some dissimilarity in them. The number should represent mutual information.
Hi Haris,
Could you give more details why you need an index of 0 or 100 when images are exactly the same? Or do you just need a scale between 0 and 100? If I understand correctly your need, you have to first consider the co-occurrence matrix between your two images (in case of color image CRCM). Then, you can compute either the Entropy, Energy or Mean value based on last mentioned co-occurence matrix. Such computing is supposed to give you an index of similarity. Then several approaches allow to normalize the index between the extreme situations as similar image and two random noisy images.
My question is that, a value of 0 or 100 should be returned if 2 images being compared are exactly the same, otherwise the returned value should be different that that difference should base on mutual information of images. I read about mutual information but the value returned by it is confusing for me.
Hi Haris,
as suggested by Hubert, I guess you need to scale the MI to the desired range. One way to do so is to calculate the IM between two identical images and between the same image and white noise. You should get then maxMI and minMI respectively. Then scaling would be trivial e.g. (x-minMI)/(maxMI-minMI)*100. Does it make sense?
Mutual information between 2 random variables X and Y defines the dependency of these variables, i.e. mutual information reveals how much X contains information about Y, and vice versa. Mutual information can be utilized to quantify the temporal correlation between frames of a video sequence, so it can calculate the amount of redundancy between any two frames.
I used this measure in a chapter about VSR, The temporal change of the appearance of ROI is caused by uttering a new/different phoneme. For example, the mouth appearance will change while switching from phoneme [ē] to phoneme [d] when uttering the word “feed”. Therefore, it is sensible to use the mutual information to measure some aspects of the change in the mouth area between consecutive ROIs.\
for more about this you may refer to : https://www.researchgate.net/publication/257934479_Visual_Speech_Recognition
Chapter Visual Speech Recognition
- Badges
- Science topic
- Similar topics
- Computer Science
- Images