|(10th September 2014)|
Nowadays there are many HTML5-based map services, but typically they don't offer any export functionality. To create a full view of the desired region, one can either zoom out (and lose map details) or take many screenshots of different locations and manually stitch them together. This project can automatically load all stored screenshots, detect the map, crop relevant regions, determine images relative offsets and generate the high-res output with zero configuration from any map service.
|(9th August 2014)|
This project's goal was to automatically and robustly estimate and compensate distortion from any receipt photos. The user is able to just snap the photo and OCR could accurately identify bought products and their prices. However this task is somewhat challenging because typically receipts tend to get crumbled and bent. Thus they won't lie nicely flat on a surface for easy analysis. This set of algorithms solves that problem and produces distortion-free thresholded images for the next OCR step.
|(7th June 2014)|
From my office window I've got an unblocked size-view to the Ring Road I (Kehä I) in Espoo, Finland. It is one of the busiest roads in Finland, having up-to 100.000 cars / day. I wanted to create a program which would receive a video feed from a webcam and would process images in real time on common hardware.
|(6th July 2013)|
My Masters of Science Thesis involved the usage of a so-called "omnidirectional camera". There are various ways of achieving 180° or even 360° view, with their distinct pros and cons. The general benefit of these alternative camera systems is that objects don't need to be tracked, because generally they stay withing the extremely broad Field of View (FoV) of the camera. This is also very beneficial in visual odometry tasks, because landmarks can be tracked for longer periods of time.
|(26th June 2013)|
I developed a coin recognition system which can recognize eight different groups of coins. The used set is all five coins of Singapore, but a few categories cannot be distinguished from each other without knowledge of the coin's size in relation to others.
|(25th June 2013)|
For my Bachelor of Science degree I developed a novel fingerprint matching algorithm, which ended up beating many alternative methods which were developed by research groups around the world. The used dataset the same which was used for FVC 2000 (Fingerprint Verification Competition).
|Platform||(About this blog)|
|Introduction to Stable Diffu...||(2022 Nov)|
|Matching puzzle pieces together||(2022 Jul)|
|Single channel speech / musi...||(2022 Feb)|
|Image and video clustering w...||(2022 Jan)|
|Helsinki Deblur Challenge 2021||(2021 Dec)|