A fine-grained classification method for women's fashion images based on part detection and visual features
A technology of visual features and classification methods, applied in computer parts, character and pattern recognition, instruments, etc., can solve the problem of not making good use of local information, and achieve the effect of improving accuracy and high classification accuracy.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0030] Embodiment 1: as Figure 1-2 As shown, a fine-grained classification method for fashion women's clothing images based on part detection and visual features, firstly detects the parts of body parts on the input fashion women's clothing images to be classified and fashion women's clothing images in the fashion women's training set; secondly, extracts parts respectively The detected fashion women's clothing image and the HOG, LBP, color histogram and edge operator four underlying features of the fashion women's clothing image are trained to obtain the image after feature extraction; then, the defined visual feature descriptor is combined with the extracted four The underlying features are matched, and the fine-grained classifier model is trained using random forest and multi-class SVM supervised learning; finally, through the trained fine-grained classifier, the fine-grained classification of the fashion women's clothing images extracted from the feature is realized, and th...
Embodiment 2
[0036] Embodiment 2: wherein the improved DPM model is made up of a root model and some part models, and the object model of n parts is represented as a (n+2) tuple (F 0 ,P 1 ,...P i ,...P n ,b), where F 0 is the root filter, P i is the model of the i-th component, b is a deviation loss coefficient, at l 0 scale layer, with (x 0 ,y 0 ) for the anchor's response score is:
[0037]
[0038] in, is the response score of the root model, v i Is a two-dimensional vector, used to specify the coordinates of the i-th filter's anchor point position (ie, the standard position when no deformation occurs) relative to the root position, is the response score of the n part model, λ is the number of levels of the feature map computed at twice the resolution in the feature pyramid;
[0039] After calculating the response score, transform the response of the component filter and take into account the spatial uncertainty, the response transformation calculation formula is as follo...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


