guangyangpku / captcha-recognition-based-on-dsp-using-machine-learning-techniques Goto Github PK

###Purpose This project is for ELEN 4810E Digital Signal Processing class @columbia university.

###Enviroment

• Operating system: MAC OS X EI Capitan
• Language: python
• Lib: PIL, requests, sklearn, numpy

###Program

####scrapyPicture.py

• A web scrawler to download the verification code from website.
• Url link : 'http://scm.sf-express.com/isc-vmi/loginmgmt/imgcode?a=0.18090831750297091'.
• Download the images as *.jpg and save them into folder /pictures.

####binary.py

• This program is used to binarilize pictures.
• This program contains two classes: class BinarizingImage is used to change the images into binary base, and class savingImage is used to save the splitted letters into folder /splitPhotos.

#####class BinarizingImage:

######functions:

• loadPicture(self, _filePath, _threshold): Load pictures.
• binarizing(self): Using threshold to binarilize the pictures.
• depoint(self): Using connected component labeling method to suppress noise.
• equidistanceSegment(self): split image in equal distance

#####class savingImage:

• save pictures as *.bmp, input an Image list.

######functions:

• saveImage(self, images): save images into folder '/splitPhotos'.

####createTrainingDataset.py

• This program is used to divide images into different folders by their label.
• After that, we need to justify the result manully, the accuray of OCR is about 80%.

####Extraction.py

• use to write the training dataset as traindataX.csv and traindatay.csv

#####class extractFeatures:

######functions:

• getBinaryPixel(self, _filePath): from *.bmp get get a binary pixel array.
• getFileNames(self, _dirs): Scanning all the file in the dirs.
• writeFile(self): Writing training dataset.

####SVMmodel.py

• This program is used to train SVM model, and use joblib to dump the model as SVM_PKL.pkl.

#####class SVMmodel:

######functions:

• loadData(self, _train_X, _train_y): from 'traindataX.csv' and 'traindatay.csv', load training dataset.
• train(self): use clf.fit(X,y) to train the model.
• selfTest(self): self test the accuary of our model.
• searchBestParameter(self): using sklearn.grid_search to find the best parameters of SVM model.

####finalPrediction.py

• This program use trained SVM model to varify a new code.

#####class predictionModel:

######functions:

• loaddata(self, filePath): from traindataX.csv and traindatay.csv, load training dataset.
• SVMpredict(self): use clf.fit(X,y) to train the model.

####KneighorsModelPrediction.py

• This program use Kmeans method to varify a new varification code.

####classification.py

• This program will use trained SVM model to verify the images in testpictures folder.