Vislog

I have no idea what happened to my domain hoster. The outage has been more than a week, and my site's still dead. I have no idea what's wrong with them but I'm dead fed-up. It's really time to get a new hoster.

My blog's stranded, not able to upload new pictures. This is so frustrating. They have pushed the wrong buttons a bit too far.

I need to get my lazy bum up to write it. But, I'm doing my literature search quite thoroughly (though I don't say I've finished "reviewing" them to a great extent). Just making sure I've gotten all bases covered.

Have been mucking around with all sorts of things. AI, OpenCV. I'll be teaching a games programming course next trimester, which involves J2ME. So, time to brush up on my Java as well.

Meanwhile, C++/OpenGL seemed to have caught my attention rather quicker than Java. Phew, no time for everything.

Nailing down a PhD scope haven't been easy. Is it too little? Is it too insignificant? Has this been done before? (Oh, the dreaded one...)

Good thing, finally sensed that there's some significant progress in the direction of work.

Andrew suggested that the "video-based face recognition" should inculcate all aspects of the true "paradigm" of a video-based framework. A tri-foci piece of work looks sensible enough:

1. Video-based face feature representation
2. Biologically inspired method (not necessary NN)
3. Incremental learning

I have been looking through Gorodnichy's paper ("Video-based Framework for Face Recognition in Video") from CRV'05 and it's quite an eye-opener.

Gorodnichy examines 3 factors on why humans are able to recognise faces in low resolution in video without much problems. This idea is indeed biologically motivated. What I see is actually a fresh new paradigm for video-based FR.

1. Detection: Very efficient mechanisms to detect a face prior to its recognition.
2. Decision: Accumulation of results over a period of time rather than one particular instance.
3. Learning: Efficient neuro-associative mechanisms.

Number 3 is probably interchangeable with other neural models, though I think Gorodnichy was brilliant in using just binarized information of various versions. Experimental evaluation didn't seemed to cover a large enough population of subjects, and varying pose and illumination conditions.

Nevertheless, a good read to start with...

Have been doing plenty of reading lately. Mostly reviews and discussions on dodgy issues involving face detection (ah, it's pose, illumination, pose, illumination...), basically "looking into the big picture". The literature is really enormous. MH Yang's ICPR2004 tutorial is really handy, and everyone should at least have a good look at it.

Video-based methods (for face detection) have the luxury of extra cues (motion, depth) and that helps narrow down the search space. I'll look into it with more detail later, and work from there on towards the 'recognition' part for my PhD.

I'll be having two final year projects on face detection for undergraduates next trimester (Nov), and hopefully, there are people interested to take up the challenge. No details for now though, till they're released!

Took a break and came upon this...



Some good mental exercise. What you have to do is to bring the gray dog (to me it's a cow) to the yellow umbrella (the goal). I'm figuring out Level 23 now ...

Have been looking into various LaTeX distributions today. Mostly packed for Linux machines except for a few that have been so kindly ported over to Windows, MikTeX (oh so popular), LyX and TeX Live (which I think is unnecessary to have a few distributions in one pack)

LyX is offering a WYSIWYM document processor using LaTeX, which should be an incredibly powerful front-end for easy typesetting. No tears! Spent quite awhile grumbling over the need to use MinGW and Msys to remake LyX from source codes.

1.3.6 does supports Windows platform! I might try spare some time to fiddle with it soon.

I've decided to do something to keep track of my research work. Need to keep myself organised. I'm trying something unconventional, like, using a blog. Paper jots sometimes get lost, though I think they're much handier.

I didn't think of any creative catchy names, so simply just, Vislog should suffice. 'Vis' for vision or computer vision to be exact, which is basically my core area of research; 'log' for logbook. More on my research work from my academic site.

I opted to use Blogger this time (much easier and quicker, less config), over my favourite Textpattern to run Vislog. Then, auto-port it over to my personal site via FTP.

Anyway, I was thinking, wouldn't it be great to have a research group blog? Perhaps, in the future. Plenty of opportunities for everyone to discuss, share stuff, troubleshoot problems, group reading, etc. All for the fun and work!