Herk-Gil-Werman(HGW) fast algorithm for morphology

This repository does NOT depend on external libraries. However, in order to run the demo, you need to have OpenCV (tested on version 2.4.8) installed. To build the demo, just run

./buildDemo.py

matWrapper.hpp defines a template MatWrapper, which serves as a light alternative of OpenCV Mat. The usage is demonstrated in the demos.

Reference

Van Herk M. A fast algorithm for local minimum and maximum filters on
rectangular and octagonal kernels[J]. Pattern Recognition Letters, 1992,
13(7): 517-521.

Demo

• compareEfficiencyWithOpenCV
• erodeImage

Directories

• data: test images
• demo: demo source code

Efficiency comparison with OpenCV

The following result is tested on 1.4 GHz Intel Core i5, with 4GB RAM and Mac OSX. Note that:

1. OpenCV uses SIMD, so it's very fast

2. Due to the fact that SIMD functions can only process a fixed number of bytes in parallel, efficiency of OpenCV implementation decreases as we process 16Bit images of the same size

3. Pixel depth has little effect on HGW implementation

    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [3 x 3]
    elapsed time of OpenCV erode: 0.021483 secs
    elapsed time of HGW erode:    0.699951 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [19 x 19]
    elapsed time of OpenCV erode: 0.02735 secs
    elapsed time of HGW erode:    0.576092 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [3 x 3]
    elapsed time of OpenCV dilate: 0.008891 secs
    elapsed time of HGW dilate:    0.610029 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [19 x 19]
    elapsed time of OpenCV dilate: 0.022744 secs
    elapsed time of HGW dilate:    0.578881 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_16U) size: [5472 x 3648]
    Kernel size:        [3 x 3]
    elapsed time of OpenCV erode: 0.049 secs
    elapsed time of HGW erode:    0.437014 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [19 x 19]
    elapsed time of OpenCV erode: 0.02735 secs
    elapsed time of HGW erode:    0.576092 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [3 x 3]
    elapsed time of OpenCV dilate: 0.008891 secs
    elapsed time of HGW dilate:    0.610029 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_8U) size: [5472 x 3648]
    Kernel size:       [19 x 19]
    elapsed time of OpenCV dilate: 0.022744 secs
    elapsed time of HGW dilate:    0.578881 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_16U) size: [5472 x 3648]
    Kernel size:        [3 x 3]
    elapsed time of OpenCV erode: 0.049 secs
    elapsed time of HGW erode:    0.437014 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_16U) size: [5472 x 3648]
    Kernel size:        [19 x 19]
    elapsed time of OpenCV erode: 0.058585 secs
    elapsed time of HGW erode:    0.367326 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_16U) size: [5472 x 3648]
    Kernel size:        [3 x 3]
    elapsed time of OpenCV dilate: 0.017879 secs
    elapsed time of HGW dilate:    0.313497 secs
    Result error  estimation: 0
    =========================================================
    Image(CV_16U) size: [5472 x 3648]
    Kernel size:        [19 x 19]
    elapsed time of OpenCV dilate: 0.101644 secs
    elapsed time of HGW dilate:    0.453101 secs
    Result error  estimation: 0