Method and system for image classification
View Patent ↗Methods and systems are disclosed for image classification coding an image by nonlinearly mapping an image descriptor to form a high-dimensional sparse vector; spatially pooling each local region to form an image-level feature vector using a probability kernel incorporating a similarity metric of local descriptors; and classifying the image.
1. A computer-implemented method for image classification, comprising:
a. coding an image by nonlinearly mapping an image descriptor to form a high-dimensional sparse vector;
b. spatially pooling each local region to form an image-level feature vector using a probability kernel incorporating a similarity metric of local descriptors; and
c. classifying the image;
wherein the nonlinear mapping of f(x) is approximated by a linear function w T φ(x), where φ(x) is called the Super-Vector (SV) coding of x, defined by
φ( x )=[ sγ v ( x ),γ v ( x )( x−v ) T ] vεC T ,
and wherein the spatially pooling forms a modified Bhattacharyya kernel.
2. The method of claim 1 , comprising representing the image as a set of descriptor vectors with their 2D location coordinates.
3. The method of claim 1 , wherein the descriptor comprises one or more local features.
4. The method of claim 1 , wherein the pooling comprises, for each local region, aggregating codes of all the descriptors to form a single vector and concatenating vectors of different regions to form an image-level feature vector.
5. The method of claim 1 , comprising performing spatial pyramid matching to incorporate spatial location information.
6. The method of claim 1 , comprising applying a linear support vector machine (SVM) to classify the image.