1## OpenCV Hierarchical Feature Selection for Efficient Image Segmentation module 2 3Author and maintainers: Yujun Shi (shiyujun1016@gmail.com), Yun Liu (nk12csly@mail.nankai.edu.cn). 4 5Hierachical Feature Selection (HFS) is a real-time system for image segmentation. It was originally proposed in [1]. Here is the original project website: http://mmcheng.net/zh/hfs/ 6 7The algorithm is executed in 3 stages. In the first stage, it obtains an over-segmented image using SLIC(simple linear iterative clustering). In the last 2 stages, it iteratively merges the over-segmented image with merging method mentioned in EGB(Efficient Graph-based Image Segmentation) and learned SVM parameters. 8 9In our implementation, we wrapped these stages into one single member function of the interface class. 10 11Since this module used cuda in some part of the implementation, it has to be compiled with cuda support 12 13For more details about the algorithm, please refer to the original paper: [1] 14 15### usage 16 17c++ interface: 18 19```c++ 20// read a image 21Mat img = imread(image_path), res; 22int _h = img.rows, _w = img.cols; 23 24// create engine 25Ptr<HfsSegment> seg = HfsSegment::create( _h, _w ); 26 27// perform segmentation 28// now "res" is a matrix of indices 29// change the second parameter to "True" to get a rgb image for "res" 30res = seg->performSegmentGpu(img, false); 31``` 32 33python interface: 34 35```python 36import cv2 37import numpy as np 38 39img = cv2.imread(image_path) 40 41# create engine 42engine = cv2.hfs.HfsSegment_create(img.shape[0], img.shape[1]) 43 44# perform segmentation 45# now "res" is a matrix of indices 46# change the second parameter to "True" to get a rgb image for "res" 47res = engine.performSegmentGpu(img, False) 48``` 49 50 51 52### Reference 53 54[1]: M. cheng, Y. Liu, Q. Hou, J. Bian, P. Torr, S. Hu, Z. Tu HFS: Hierarchical Feature Selection for Efficient Image Segmentation ECCV, Oct.2016.