devashishprasad/embossed-text-reader

This is a tesseract based OCR to read Embossed text on metals. It can also be use as a general OCR.

👋 Hi there

🧬 I am currently harnessing generative AI to address the most complex challenges in Biology for advancements in drug discovery.

🌟 I have published 3 research papers in deep learning /machine learning. One of the papers was published at CVPR and has over 150 citations and 1.3K Github stars.

🎓 In the summer of 2023, I earned my Master of Science in Computer Science, specializing in Machine Learning, from Purdue University, achieving a GPA of 3.6/4.0.

🖥️ During my final year of graduate studies, I served as a Research Assistant at Kihara Lab, one of Purdue University's premier applied ML research facilities. There, I employed generative deep learning models for intricate protein structure analysis and established a machine learning serving infrastructure for these computation-intensive models (em.kiharalab.org).

👻 In the summer of 2022, I joined Snap Inc.'s Camera Platform team as a Machine Learning Engineer intern. My role involved developing and deploying a deep learning-based optical flow prediction model to automate Snapchat's video annotation process. This initiative enhanced video labeling and annotation efficiency by 15% for Snapchat's gigantic unlabelled video datasets.

🛰️ In my first year of graduate study, I collaborated with Viasat Inc. through Purdue's The Data Mine program as a Graduate Data Science Researcher. My research focused on developing Deep Learning algorithms to tackle blind image super-resolution challenges, intended specifically to enhance the quality of Viasat's internal satellite imagery.

🕸️ Before my relocation to the US, I completed five internships in India, focusing on Machine Learning and Deep Learning in the topics of News sentiment analysis, ML in finance, Sports vision analysis, Document understanding, Optical Character Recognition, Face Recognition, Fine-grained image classification, Chatbots, etc.

🏅 I was the Smart India Hackathon (India's biggest Hackathon) grand finalist three times. During which, I worked on ML/DL-based projects for ISRO, ITC Ltd, and DRDO (India's esteemed organizations).

www.devashishprasad.com

embossed-text-reader's Issues

Not working video webcam

VIDEOIO ERROR: V4L2: Could not obtain specifics of capture window. VIDEOIO ERROR: V4L: can't open camera by index 1 /dev/video1 does not support memory mapping Traceback (most recent call last): File "video_from_webam.py", line 12, in <module> img = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) cv2.error: OpenCV(4.0.0) /home/[user]/opencv/modules/imgproc/src/color.cpp:181: error: (-215:Assertion failed) !_src.empty() in function 'cvtColor'

Not working on given images

@DevashishPrasad
I tried on following image but it does not gives expected output .

Can you please help me on these ?
Also which research paper r u referging?

Error on ContourAreas

Hi,

While initially playing your code i was running into this error:

`
File "C:/Users/thinkpad/Desktop/read_serial.py", line 56, in
if cv2.contourArea(c) < 300:

error: OpenCV(4.0.1-dev) C:\Users\thinkpad\Documents\openCV4Win\opencv-master\opencv\modules\imgproc\src\shapedescr.cpp:274: error: (-215:Assertion failed) npoints >= 0 && (depth == CV_32F || depth == CV_32S) in function 'cv::contourArea'`

I fixed it by adding one variable in this line:

#cnts = cv2.findContours(dilate.copy(), cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE) cnts,hierarchy = cv2.findContours(edged.copy(), cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)

It seens related to how different versions of OpenCv need different numbers of parameters for findContours.

I would have thought that the next line would cover for that, but apparently not:
cnts = cnts[0] if imutils.is_cv2() else cnts[1]

Anyway, I got it working, and I actually post this issue because I didn't find another way to thank you and send my greetings :)
Cheers!