Spatial analysis of salient feature points has been shown to be promising in image analysis and classiﬁcation.
1. Bag-of-features (BoF)
- BOF treats an image as a collection of unordered appearance descriptors extracted from local patches;
- Then the patches or descriptors are quantized into discrete visual words of a codebook dictionary, and
- then the image histograms are compared and classiﬁed according to the dictionary
BUT it discards the spatial order of local descriptors
2. Spatial Pyramid Matching (in scale space): extension of the BoF model
To match salient feature points + spatial multiresolution blocks in multiple scales to reﬁne the kernel for SVM classiﬁcation.
BUT, different blocks may have similar feature
people can recognize natural scenes while overlooking most of the details in it (i.e. the constituent objects).
Scale is an important aspect of local feature ﬁnding in prominent cue detection in images. E.g. SIFT, which uses the maxima/minima of neighboring scale space to ﬁnd the interest points or key points of an image.
3. Scale-space theory
It present images as one parameter family of smoothed image, according to the size of the smoothing kernel used for suppressing ﬁne-scale structures
4. Subdivide and Disorder principle
The essence of this principle is to partition the image into smaller blocks and calculate orderless statistics of low level image features (e.g., pixel value, gradient orientation, filter bank output etc.).
The subdivision method:
- regular grid,
- quad trees, and
- ﬂexible image windows