I can never remember faces. I know that I have met someone, but can never place her or him. Did I meet him at some cocktail party last year? Did I meet her when I went to watch a play? I can never get these things straight. As a result, the first few minutes of my conversation with people I should know is always awkward! Many a times, after a particularly awkward conversation with someone who turned out to be my fifth standard best friend, I have wished for an application which will tell me who someone is just from her or his picture.
Thus, I was naturally intrigued by the application that won the SAP d-code 2014 Bangalore Demo Jam on March 13: It did exactly what I wanted! On entering a picture of a person, the application finds other pictures of the same person, and even relevant information. I had to know more about this. Thus, I interviewed the developers behind this software. Here is the interview, in full.
1. Hi, I loved the application you demonstrated at the Demo Jam. If I understand it correctly, it’s am image database search. Can you tell me more about it?
We’ve tried to focus on the intersection between image processing and big data. There are some problems which image processing can solve, like object detection, recognition etc. and there are some problems which come under big data. Image processing can get slow and problematic when dealing in large amounts of images.
So when looked at this way, HANA looked poised to be the solution. Then it was just a matter of putting the idea to code. Our implementation allows you to search for similar images based on an original histogram matching algorithm. Not only that, but it even recognizes faces and can then provide relevant information. It can do all of this with massive amounts of data.
2. How did you have the idea of working on such an application?
It all started on Annual day 2013 at SAP Labs India, Pune. Anirban Dey, MD, SAP Labs India, had attended the event. He noted that there are currently no image processing apps on HANA and wanted some engineers from the Pune Labs to take up the challenge. We did!
3. So how long have you been working on this idea?
We started to tinker around with the concept in January of 2014. This is not our day-to-day work, so we mainly worked on this at nights or on weekends. Initially we came up with a simple algorithm which when given an image, found similar images. We kept improving it till it could work for face recognition. Then came the scaling part. We had already tried with about 3000 images, but we wanted to really put the system to the test. So we kept challenging it till we hit a mark of 8.2 million images. Searching an image data set of that size… it felt good.
4. What are your next steps after winning the contest?
We’re still working to improve it! It’s a work in progress! We participated because we wanted to showcase the image processing capabilities of HANA.
5. How was the experience of participating in the Demo Jam and trying to get your idea across in six minutes?
Frankly, it was an out-of-the-body experience. We were here for the first time…and winning it was definitely not on our mind. We just wanted to enjoy the event and absorb the terrific atmosphere!