Celine:
This week, my groupmates and I wrote up our design paper together. I was sick for the most of this week, so have not been able to make significant progress. The progress that I did make was testing out text detection using a pre-trained EAST text detector convolutional neural network, in order to try to segment text from page sets with images. The results were not what I wanted though, so for this problem I will need to try something else. For now, the data I have tried so far has worked well using pytesseract, as it will ignore the illustration on the page (please see the second figure below).
I am in the middle of trying to implement some text/page skew correction to try to improve the output from using pytesseract. When I input an image that is clean and not skewed, with high resolution, pytesseract works very well:
However, when used on an actual image that I took with my phone, I get results like this (see linked image for better resolution):
I noticed that “Stuart” is recognized incorrectly when it is along the curved part of the page, but recognized correctly when it is on the flatter portion of the page, so I am hoping that some skew correction will improve recognition.
I hope to have this skew correction completed this weekend, and in the coming week will implement some image processing such as binarization to see if that will improve accuracy as well. During this coming week I hope to have a working python script that takes an image of a page set against a black background and have it perform with better accuracy than it does in the image shown above.
Effie:
Indu:
This week I worked more on the design of the page turning device, specifically how the wheel and gear should connect to the motor in order for it all to work for turning the page, by having drawings of the potential design being that it would include a pivoting mechanism in order to allow it to be lifted up when the page is being turned.
This Wednesday, our Raspberry Pi came in so Effie and I spent majority of class time, setting it up and further discussing how we think the Pi will be used to operate all the different components of the device, as it will involve both of us to integrate the mechanics of the device with the Pi. Also Effie went to the library and got us our test base for the books. We all spent the rest of the week trying to use the Arducam with the Pi in order to take pictures of the books, but kept getting that the camera was undetected, so we think the Arducam may be faulty. Celine contacted Arducam to ask about the issue we were having so hopefully we get a helpful response soon.
In terms of next steps, next week I will work on building a mock version of the stand and the page-turning device so that Effie and I can connect various parts of our device (e.g. the wheel for page-separating, the gear for page-turning) to the Pi in an attempt to make each part work individually. We would also like to test other page-turning methods, as we stated earlier that while we think the conveyor chain method is the gentlest, we want to test this to actually know for sure.