What is the most efficient descriptor for analyzing and describing image textures in terms of computational complexity, sensitivity to rotation, ability to combine with color information, gray-scale sensitivity, etc?
LBP, CLBP, SIFT, GLCM, oRGB-LBP, CLCM, CCTD, LTP, HCLBP, NrLBP, etc?
Please explain your idea? Which descriptor, Which application?
We can point some disadvantages for basic LBP as follows:
1- High sensitivity to noise
2- Features with low discrimination
I prefer some new versions of LBP such as :
Hybrid Color Local Binary Patterns (HCLBP)
Local Ternary Patterns (LTP)
Completed LBP (CLBP)
Improved Local Quianry Patterns (ILQP)
Hi,
i recommend the deep learning model such as Convolutional Neural Networks, they outperform all the existing descriptors.
Hello, yes off course deep learning and machine learning exist with each other.
deep learning is best to use.
I think, first the term 'efficient' should be described. Which one of the following aims can describe 'efficient' better?
- providing most discriminant features
- low computational complexity
- rotation invariant
- noise invariant
- gray scale invariant
-etc
I do not think that a descriptor can be proposed to satisfy all of the above objectives at the same time.
Dear Shervan The proposed method which mentioned below, has a significant performance for the gray scale textures in terms of rotation, different noises (with several distributions), computational time and complexity, gray-scale sensitivity.
Article A new approach for rotation-invariant and noise-resistant te...
There are some interesting aspects and challenges in this paper and method as follows: 1. An integrated method without changing the initializing and settings parameters and number of features to obtain significant results for different datasets (fixed initialization). 2. No need to a lot of extracted features and many training data for each class that result low computational cost (few train textures and only two simple features; mean and standard deviation). 3. More practical and applicable method with suitable low computational time and complexity (almost real time). 4. An efficient and resistant method against different noise types with different distributions (Five different noise types, e.g. Gaussian, Salt&Pepper,Periodic,Speckle and Poisson). 5. Utilizing many texture images for evaluation of proposed method (more than 500000 test textures).There are also acceptable evaluations and comparisons in this paper. I suggest to other friends to read and evaluate this method or compare with other method, specially in terms of different noises and other challenges mentioned in this paper. I believe that this method may also have a good performance for color images, because the method was evaluated for OUTEX dataset which has some color information, but it is not evaluated for color images yet.
Remove Noise in the images
Remove Shades in the images
first convert image in gray scale and convert in binary image and then indexed image to perform algorithm
A complete survey about texture image descriptors can be found in the following link:
https://www.researchgate.net/publication/332370066_TEXTURE_IMAGE_ANALYSIS_AND_TEXTURE_CLASSIFICATION_METHODS_A_SURVEY
i want to present my paper of matlab in computer science conference please let me know ? Shervan Fekri Ershad
- Badges
- Science topic