1. To detect the regions of characters from a natural image.

2. Then the detected regions are filtered reduce the false postives and then it is feed to the Neural Network which classifies each image to its corresponding characters.

3. Then these characters are then printed on the console.

1. First, we apply the canny filed to highlight the edges of the natural scene.

2. The highlighted edges are then used to highlight the contours which are then extracted with the help of OpenCV.

3. The contours are then passed to the text detection algorithm ( SWT, MSER ,ER ) and then the if the contours are a text, they are marked. And to reduce the number of redundant number of detected regions overlapping regions are grouped together and merged.

4. The marked masks are then feed to ConvNet. The Neural Network is trained with fonts and handwritted characters from the Chars74K dataset. Then this ConvNet is used to evaluate the mask and if the mask is classified into a letter, then the characters are printed in the console.

1. I tried SWT from scratch but the performance was too bad, so had to go for the vannila version of NM classifier used in OpenCV.

2. Implemented the Tessaract-OCR version for character identification to mesure the performance achieved so far, but in the next release will go with a full fledged Neural Network using TensorFlow.

1. I have used the Opencv implementation of the algorithm refered in the paper "Real-Time Scene Text Localization and Recognition" by Luka ́sˇ Neumann and Jiˇr ́ı Matas, to detect regions of text present in a image.

2. Then the regions are extracted into a mask and then it is fed to the tessaract ocr for robust identification of alphanumerical charecters in the natural scene image.

3. The extracted characters are then displayed on the console.

1.Background subtraction.
2.Update every 5 seconds .
3.Multithread the TTS Engine to simultaniusly track the image as well as the Speech
4. if possible import an acclarometer and a gyroscope to detect motion,


TEXT EXTRACTION:

1. First, we apply the canny filed to highlight the edges of the natural scene.
2. The highlighted edges are then used to highlight the contours which are then extracted with the help of OpenCV.
3. The contours are then passed to the text detection algorithm ( SWT, MSER ,ER ) and then the if the contours are a text, they are marked.
4. ##################
5. And to reduce the number of redundant number of detected regions overlapping regions are grouped together and merged.
6. ##################
7. Pass the masks into DS where the points are sorted In ascending order based on the height and Width (just the point data ),
8. Iterate thought the DS and sent the conouturs point to the Convnet.
9. The marked masks are then feed to ConvNet. The Neural Network is trained with fonts and handwritted characters from the Chars74K dataset. Then this ConvNet is used to evaluate the mask and if the mask is classified into a letter, then the characters are printed in the console.

Then use google TTS or ….. offlline tts

use page number to check if same page

use the ultrasonic sensor to check if book is within proximity (every book has equal fonts and at particular distance and position must be told the AI to user via book edge and distance)