 Good afternoon everyone. I am Saurabh Kumar Jindwani from Ayeshaan Munbat and I am working on image assessment tool set for the drawing course which has set up on EDX. As Bandit, my team mate told you all before, this whole group is a research based project on things that have never been done before. So for our initial goal we have chosen simpler scope for our projects so that we can achieve it within the time limit. My tool will take a teacher's drawing which is a correct image and a student's drawing and assess it for matching. This is my application overview. First I made a course on EDX for drawing which will be used by students to learn it and they have got assignments to draw and to assess themselves. The open source libraries I used are OpenCV and Python for the image matching. OpenCV is open computer vision which is image processing library available and for the front end I used HTML, PHP and JavaScript. The inputs to my application are the user's image and the correct image. For pre-processing I am rectifying the lightening variations as the user may click it with his own camera. So the lightening might vary. Noise reduction and I am converting it to a binary image and also just as I said I am also taking care of the size of the image and position. For matching I am breaking down the image into small individual components and then I am comparing the components of the correct image with the corresponding components of the user's image. And as the response I give an over-a-score and a visual feedback. I will show it now. Is it painting or drawing? Drawing. Only drawing. Only using what? Only any kind of pencil. Yes. So that is. They have to, the student will have to draw something like this and upload. So you draw it. So you draw it. No, you draw it. I have a drawing drawn image. I love it. You draw, yeah. How difficult is it to draw? Can you draw it? Yes, I will draw it. Here draw it, take a picture. You cannot draw it. Oh, you have taken a picture. Yes, I will. Upload of a picture. It's not a drawing tool. No, what I am saying is that this method is, it looks like it will just check with it the Xerox copy of this image. No, no, no. You upload something. Yes, sir. Yes, he is skipping the video. Let's see. Like this is the image I have drawn with my hand. It has clicked with my photo camera. So it has some lighting. The previous image is drawn by what? Who has drawn that image previously? User with pen or pencil on paper. It's not exact. This is very exact. The original one is so exact. The original one is from where? The original course I had taken from the web. What license? Not license. It was SIN by ghost. So this is the correct image and user uploads some image like this. Then we will calculate the score. It will be something like this. You can see that user has got 1, 3 or 10. And some visual feedback on what part he has drawn wrong or correctly. Like this. The red parts are showing the wrongly drawn parts. The orange parts are intervening correctly. And the green part which you can see is totally correct. So the scoring is based on? But there is a difference between the green. There is a big difference. The curvature if you see on the right. It's not only on the shape. It's based on various features such as shape, area, its location, its position with respect to the correct one. And I have allocated different percentages to it. So how much the student will get here? How much student will get? Which is red? We will see relatively. Which is the red? This is the original. Left one is original. Which is the red zone? Red zone. This is also? No sir. It's shape or? Shape, area and position. I don't tell you what we have done. You have to show the software if you want. Okay. Dude, he can't explain. I can actually do. Why this is red? Why did you orient it? This is the library function. For matching shapes, I am using the library function. Which? Which library? Which? It gives you a number between 0 and 1. Which library is that? Which library? OpenCV. OpenCV. That is my contour shape matching. He has calculated the two areas. Okay. So he knows how much percentage difference the area is. Third parameter is? 7.03. How much location? It is wrong. If you draw the same shape and draw it somewhere else. Okay. He will see you two third of the marks. Okay. The last part if it is showing orange, it should be red if you see. Go to the last, last right. Right mouse. This one. Correct. So I have decided. Go to the right. Correct. This one. Yes. Now what would have happened here is, what would have happened here? If you see the contour matching algorithm. Would have given you? Almost, almost exact match. Okay. But the area is completely different. I am allotting 50% of marks. The three dimensions for all these three variations. He is evaluating based on three. Okay. One is whether the space matches. Okay. Whether the area matches, where the relative position is. The different dimensions for each. We probably how much weight is for contour? For the shape there is 50% weightage. For area there is 25 and for the position it is 25. And then I am calculating the overall score. So contour it has matched and location it has matched. Okay. Area does not match. So 25% less weight is wrong. That will explain when it is red and when it is orange. Okay. So I have calculated scores for each of the components. Each of the components are adding it up. And then finally I am adding it up as the ratio of their area of the original one. There is one more thing which I don't know. You have done a weightage before this. Yes. The weightage of the area of the original one. About the scaling thing. Scaling thing. I have taken care of scaling. The original image was different. You can show it. The whole area is different. See. This was the original image. So it is a percentage match. What? It is a percentage match. Yes. You can see that. I have calculated the score out of 10. You can see this. I have taken the camera from the camera. So this is taken. The whole match is done. Yes. That was the final score. For each component I have calculated scores were 50%. Shea was also 25%. 25% in India, 20% in India. Okay. For total score. I am calculating for each component a ratio which is... Total area. For total area? Yes. Yes. That is the first scaling. See. That drawing is so small. It is already expanded. Yes. Yes. Yes. That too. Total area matches and expands. That is in the top. Total area. Yes. That is matching it too. Yes. He does the cropping. He cropped both the images and the matches. Okay. He cropped the image over there and cropped over here. That cropping he adds to this and then starts. Okay. What he has not done. He has not done the rotation. The drastic rotation we have not taken care of it. We will take next time it can be proved. Okay. So these are the works I done. First was the installation and hands-on experience with OpenCvLibrary. I am just writing this because I had to install it on each and every computer of our group. So it took a lot of time. So I had to write it. Second one is image processing for conversion of user's image to an ideal input. User's image will have noise, variable writing, variable sizing. I have taken care of that. Then I am extracting features which are known as contours which are set of points which mark an area. Exactly. Which are used in attributes of each contours to match. And then finally matching the corresponding features of the correct image with the user image and generating an overall score. I have also created a web interface for the application as the application was purely python application. So I have also created a web interface and also integrated with edX platform. So it can be used there. And also I have created drawing course for the application to use. Then the challenges I faced. First was the fine-taining of parameters for the application. Because I will tell you whatever you do, blah blah. It has got absolutely zero meaning. Unless 110 people here actually use your thing and like Aparna ma'am came. Okay. And said that I have given this. Why I have received six marks? I want to know how good your software is in the minds of the user. When I do a drawing of that cab. I think I have done it to a reasonable extent. If everybody says you have given half the mark then you have to modify your software. Okay. Because like he said all these things you have made the first attempted assessment. I don't know whether it is correct or not. We have to fine-tune it. Okay. And I cannot keep fine-tuning it unless we have data. You have done most of the work. What is required is fine-tuning it post testing. For that testing is required. Yes. Okay. Test data is required. You people should give test data. So I want to know the average of this class of 110. How many drawing teacher has given? I am done. Okay. I will pass on to her.