Sie sind auf Seite 1von 17

Modules List : 1. Preprocessing Module - Completed a. GrayScale conversion - Completed b. Binary conversion using threshold - completed c.

Invertion - Completed d. Frame extraction using Contours - Completed 2. Text Extraction Module a. Balloon detection using CCL(connected component labelling algorithm) Completed b. Balloon segmentation using grabcut or watershed algorithm Completed(only manual process. Automatic segmenting not yet completed) c. Tesseract OCR- to process text - Completed 3. Training Module a. Features extraction process Completed for only tintin characters b. Samples creation process Completed for only tintin characters c. Haar Training process Completed for only tintin characters 4. Automatic Image Tagging Process a. Object detection using cascade classifiers completed for tintin characters b. Segmenting detected objects using grabcut algorithm Completed(Manual process only. Automatic segmentation not yet completed) c. Features comparison with database Not yet completed d. Validation for automatic tagging not yet completed e. Label object if it meets certain threshold not yet completed 5. Automatic training data downloader a. Web crawler to download comic images using Scrapy Partially completed b. SVM classifier to classify downloaded data Not yet completed c. Selection process to add to training data set if it matches certain threshold not yet completed Screenshots

1. Preprocessing Module a. Gray Scale conversion Module

First module is created using OpenCV with GUI as Qt creator. First we have to load comic strip as input

Loading Comic strip

Comic strip is loaded and displayed

Next step is converting to grayscale and binary

Next step involves inverting binary image and applying frame extraction process In Frame extraction process, first contour for binary inverted image is found and then rectangular regions are selected using contour points and vectors

Each frame are separate and stored in same location

Module 2

Text Extraction process Balloon is detected using CCL Algorithm

balloon Segmentation:

Tesseract OCR Process

M o d u l e 3 T r a i ning Process

First positive samples(objects of interest are loaded) Next negative samples(images containing non objects are loaded)

Samples are created using positive and negative samples

Samples created

Next is training process

This process involves haar training process Parameters are

1. Data directory to store cascade files 2. Bg negative files information 3. numPos no of positive samples 4. numNeg no of negative samples 5. w h width and height of samples 6. mode ALL mode

7. Algorithm GAB algo ( we can also use LBP or AdaBoost to extract features) 8. numStages how many stages a classifier to process( here 20)

Module 4:

Automatic Tagging part

Object Detection part

At the end of module 3 i.e. training process, a xml file is generated for the particular object. Using that xml file (here tintin.xml ) we will detect objects in comic.

Remaining Modules yet to be completed

Module 5: Web crawling Part

Not yet completed Coding part

Das könnte Ihnen auch gefallen