EmotionDetection

This is final project for graduate, and it's only half part of whole completed project

Description

This project can detect video or stream, there are up to 7 emotion can be classfied

Steps

1. First step is trying to get as many image data as we can
2. Normalize all the image data to the same size and grayscale
3. Training the model based on preprocessed image data
4. Predict outcome on video or stream

Function

Image crawl

1. Execute imageScraper.py and you can type in query string to search on google chrome, multiple query string is seperated by space
2. Execute getUrls.js to scanf through the google image page and get all of ORIGINAL (not compressed) image urls, and then links will be saved as txt
3. Execute imageDownloader.py to download parallelly all image based on the previous txt file, the output will store in a dir

Image Preprocess

1. Execute faceCutter.py for cut out the face part parallelly, and it will save as a grayscale image
2. Execute manualClassifier.py to select a dir to classify images into 7 different categories
3. Execute dataAugmentation.py to rotate, flip, shear to increment our image dataset

Training

1. Execute emotionTrain.py to set up a model and train, the model's struture is in emotionNetwork.py

Dectection

1. Execute realTimeEmotionDetection.py for stream like web cam or other, this program will skip some frame to increase performance
2. Execute videoEmotionDetection.py for video, this program is parallel and slightly complicated structure that use semaphor, thread and lock
3. Execute estimate.py to see the overall accuracy of the result

Demo

Raw image

Processed image

Result image