Chris Pollett > Students >
Riti

    ( Print View)

    [Bio]

    [Blog]

    [CS297Proposal]

    [Deliverable 1]

    [Languages-Translation-Presentation-PDF]

    [Deliverable 2]

    [Text-Detection-Images-PDF]

    [Deliverable 3]

    [Deliverable 4 : Dataset]

    [CS297 Report - PDF]

    [CS298 Proposal]

    [CS298 Report - PDF]

Deliverable 2: Detect the text in images using openCV and machine learning

The purpose of the project is to get familiar with image processing and openCV.

IMPLEMENTATION DETAILS

Implemented using openCV and python.

DATASET

Gathered the dataset by taking snapshots from google.

dataset for reference :
1.
Images data set

STEPS USED

1. Specify the probability threshold above which should be considered text (specified as 0.5)
2. Resize the image
3. Image preprocessing (scaling, noise removal, mean subtraction)
4. Get the probability and associated boundaries with the text in the image.
5. Ignore the boundaries that have probability values less than the threshold specified.
6. Eliminate the overlapping boundaries within a certain overlap threshold (0.5)
7. Draw boundaries around the text based on above information

RUNNING THE CODE

python text_detection.py --image images/login.jpg --east frozen_east_text_detection.pb

GITHUB : CODE LINK

REFERENCES


[1] X. Zhou, C. Yao, H. Wen, Y. Wang, S. Zhou, W. He, and J. Liang: EAST: An Efficient and Accurate Scene Text Detector. Megvii Technology Inc., Beijing, China