|
Feature extraction technique | Advantages | Limitations |
|
Contourlet transform (CT) | (1) Proper for two-dimensional images processing. | Not proper for image coding because of redundant transform |
(2) More directions is used in the transformation |
(3) Able to remove noises in the smooth areas and along the borders of image in a very good way |
|
Local features | (1) The texture in image areas are shown | (1) Key-points distinguishing is required |
(2) Invariant to scale, rotation, and other transformations | (2) Comparing images may be more difficult because of the differing numbers of key-points images. |
(3) No spatial information |
|
Global features | (1) Charachtarize the whole image | Sensitive to clutter and occlusion. |
(2) The descriptors of shape and texture are classified into this group of features |
(3) Very compact images representations are shown, in which every point in a high-dimensional space of features represent an image. |
Structural features | Able to encode some information about the structure of the objects | Suitable with binary images only |
|
Histogram of oriented gradient (HOG) | (1) Both shape and texture are shown | (1) Produce a very big feature vectors resulting in large costs of storage |
(2) Suatable for objects in detecting when image is processed | (2) Cannot deal with scale and rotation |
(3) Mainlly used for objects classification | (3) Time consuming when it extract features |
|
Statistical features | (1) Easily detected as compared with structural features. | Suitable only with gray-level and color images |
(2) Not influenced very much by noises or distortions as compared to statistical features. |
|
Gray level Co-occurrence matrices (GLCM) | (1) It is a measurement of the various combination of brightness value pixels in an image. | Feature calculation is a time-wasting process in GLCM |
(2) GLCM features is direction independent because it can be gained for a one orientation as well as merging all the orientation together |
|