digits-recognition

This is the code for number(0-9) recognition in image by using SVM Classifier and HOG descripter. And the main algorithm was packed as classes in imgDistis.cpp

how to combile

mkdir build
cd build
cmake ..
make

how to use example

there need three steps for running this code.

1. save number image to a directory and Format("img%05d_d.png") should be used as image name, the second 'd' in the format must be (0-9) asscociated with number's image or 11 for noise image. for example: ./saveNUMImgs .../images-path Note. the zero should be named (0-9 ,11) asscociated with the image for training. ("img00001_0.png-> img00001_(0-9,11).png")

2. using the number-images to train number's-model.xml for example: ./train .../number-images-path

3. using the trained number's-model.xml to detect new image for example: ./detectNUMs .../new-image (.../number's-model.xml, can be ignored, because there is one in trainData directory)

exmaple

1. ./saveNUMImgs ./character_0_1_2_3/
   note. Press 's' key for saving numbers' images to (./trainData/) directory.

2. ./train ./trainData/
   note. Before run this commond, make sure you have changed The Zero in image name to associated number.

3. ./detectNUMs ./new-image ./trainData/number_model.xml
   note. The second param can be ignored if there is a number_model.xml file in trainData directory. For sample use: ./detectNUMs ./new-image

currently

there is a partial trained images model. For test, one can just run the third step (./detectNUMs ./new-image).

Notation

By the way, this project just fit our purpose, One must rewrite the segment(..) function for your own project to generate segmented numbered images for trainnig.