Hello
I have an 256*256 image and after feature extraction with gabor filter, I have an 256*256 feature matrix. But this should be a vector not a matrix to be learned.
So for a database, there are more than 1000 image with this dimensions.
How to convert this matrix to vector, that labeling be possible at the end of the row or in last column?
Thank you
Hi Seyed,
supposing that the feature matrices can be considered as grayscale images (only one channel, i.e., one value for each pixel), and that the original images are not congruent, you could unfold the feature matrix to a vector and then calculate the frequency distribution curve of this vector. Each frequency distribution curve can be then considered as a row vector, and your dataset of 1000+ images turned to a matrix with 1000+ rows.
I already did it on RGB images, see e.g. the following paper and therein cited references:
A. Ulrici, G. Foca, M.C. Ielo, L.A. Volpelli, D.P. Lo Fiego, Automated identification and visualization of food defects using RGB imaging: Application to the detection of red skin defect of raw hams, Innov. Food Sci. Emerg. Technol., 2012, 16, 417-426, doi: 10.1016/j.ifset.2012.09.008
Hope this fits your needs.
Best regards,
Alessandro
Thank you Alessandro Ulrici
Unfolding matrix to a vector and calculating frequency distribution curve of the vector was good idea, and paper was useful.
Most Regards.
Thank you Jure Čular
But with reshaping final matrix will be so big like 1*6536. I mean something like 1*600 or 1*1000.
But thanks
I think you should use some techniques like PCA or more suitable ones depending on your needs, because you need data reduction without data loss. Hope this helps.
Hi Seyed, as Brahim mentioned, PCA would be a good tool for the purpose of dimension reduction, as would be LDA and ICA. The three differ in that PCA looks for the dimensions that preserve the most information (variance), LDA seeks the dimensions that best separate the features and ICA is best for finding independent and non-Gaussian components.
Thank you Zhicheng Cao
Yes, I will use PCA, LDA and ICA to see what will happen in final classification results. Thanks for details.
Best
Hi .You can use Bag of Words , The technique that we use in text mining ,but in this area using clustering.
- Badges
- Science topic