It is difficult to say a priori, it depends a great deal on the specific applications and a number of other factors, among them:

- The type of images (fine/coarse texture, degree of similarity between the classes);

- The imaging conditions; in particular if there are changes in illumination, rotation, scale and/or viewpoint.

You could also have a look at this recent study, which is however not specific to medical images: Article Comparative Evaluation of Hand-Crafted Image Descriptors vs....

Best,

f.

Friend, it depends on the problem and the type of image. for example, analyzing a pulmonary nodule can be done by texture and shape. Then it is necessary to analyze if the image is x-ray or computed tomography. Perhaps the techniques used for one case do not suit another case.

If you specify your problem better, it may be easier to suggest a technique.

Certainly, the proper operator is different for each case. But assuming there is not much time to examine a large number of operators, would you prefer texture analysis descriptors, color analysis operators or combination operators first?

Friend, repeating .. depends on the problem and the type of image. it is difficult to say which is the best solution without knowing the problem.

Best question about your application is: What are your images? Also we can express our question in another way. What are your objects in your image? When you answer these questions, then you can discuss the image descriptor.

I suggest the powerful descriptors such as he scale-invariant feature transform (SIFT), Pseudo-Zernike moments, Wavelet transform, gray-level co-occurrence matrix (GLCM), and other texture descriptors.

Does not depend on the image but depends on what you are looking for !!! For texture analysis then LBPs are good enough in terms of GLCM or other algorithms. If shape description then you need chain code or other spatial descriptors. Times you will need multiple depends on your operators and algorithm.

Please follow discussion for texture image classification using the following link:

https://www.researchgate.net/post/The_most_efficient_image_texture_descriptor

As some friends said, it really depends on the problem. For instance, if you want to recognize compression fracture patterns, you could use shape-based features. For histological lesion patterns, you could use texture-based features. For tumoral patterns, you could use higher-order texture (wavelet) features. And so on...

Regards

Dear Shervan

It is so difficult to say what method is better. As you know, it is depend on several concepts and situations. However, the proposed method which mentioned below, has a significant performance for the gray scale textures.

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.

2. No need to a lot of extracted features and many training

data for each class that result low computational cost.

3. More practical and applicable method with suitable low

computational time and complexity.

4. An efficient and resistant method against different noise

types with different distributions.

5. Utilizing many texture images for evaluation of proposed

method.

There are also acceptable evaluations and comparisons in this paper. I suggest you to read this paper and method.

It depends. For instance, in computer-aieded diagnosis, the prediction performance is closely related to image descriptor and machine learning classifier, and thus, feature selection and parameter tuning become important.

In general, it involves a lot of work.