Stuart James ( Tooey )

To start with my aim is to integrate Windows Phone 7 and Android with my current retrieval. This will allow me to explore how users sketch on phones as well as make a good looking demo!

So to start with Windows Phone 7, I will post a blog on the development of my completed interface and the code to make it later but I have got a basic doodle pad running that is quite a satisfying start for my first day of Windows Phone 7 development.

I had an idea for using GrabCut in some of my work and wanted to give it a quick go so went online and downloaded GrabCut source code from http://www.cs.cmu.edu/~mohitg/segmentation.htm thats an implementation of the code from Microsoft research (http://research.microsoft.com/apps/pubs/default.aspx?id=67890).

Well on there sample image:

Left is original image, right is the grabcut output

Now yets try that on some more difficult more detailed images of similar scene object... a horse

So on the left is the original image, and the middle is the generated image, with a box around only the horse and finally on the right is with the horse and the person selected within the box.

Well as you can see the results 'kind of' work but this is the best of the experiments i ran on the video frames not sure they are good enough for my current idea...

Got sent a link to this school looks really interesting anyone interested in multimedia processing should see

http://www.multimediasignalprocessing.com

Academic schedule is really impressive:

 

11 April, 2011	Theme: Computational Signal Processing
	Statistical and Probabilistic Foundations	Prof. Richard Everson University of Exeter, UK
	Principal and Independent Components  Analysis	Prof. Richard Everson University of Exeter, UK
	Multimodal Source Separation	Prof. Jonathan Chambers Loughborough University,UK
	Multi-scale Signal Processing	Prof. Fionn Murtagh Science Foundation, Ireland
12 April, 2011	Theme: Signal and Object Tracking
	Particle Filters	Prof. Simon Godsill University of Cambridge, UK
	Hidden Markov Models	Dr. Philip Jackson University of Surrey, UK
	Video Tracking: An Overview	Prof. Andrea Cavallaro Queen Mary University of London, UK
	Object Localisation and Tracking	Dr. Ian Reid University of Oxford, UK
13 April, 2011	Exhibition at Ford College
	Plenary speech: Security Imaging and Printing	Dr. Stephen Pollard Hewlett Packard Research Labs, UK
	Plenary speech: Convolutional Neural Networks for Facial Image Analysis	Dr. Christophe Garcia France Telecom, France
	Exhibition of commercial products and services
	Academic research, poster presentation
14 April, 2011	Theme: Machine Vision
	Low Level Vision	Prof. Roy Davies Royal Holloway, UK
	3D Imaging and Interpretation	(to be confirmed)
	Embedded Vision Systems	Apical Imaging
	Machine Vision Toolkits	Mr. Chris Baldwin Stemmer Imaging
	Machine Vision Toolkits	Dr. Julie Busby Multipix Imaging
15 April, 2011	Theme: Image and Video Understanding
	Colour Perception and Image Analysis	Prof. Maria Petrou Imperial College, UK
	Object Detection and Recognition	Prof. Andrew Zisserman University of Oxford, UK
	Multiple View Geometry	Prof. Marc Pollefeys ETH, Switzerland
	3D Localisation and Mapping	Dr. Andrew Calway University of Bristol, UK
16 July, 2011	Theme: Multimedia Data Management
	Multimedia Indexing and Retrieval	Prof. Stefan Ruger Open University, UK
	Video Coding	Prof. Mohammed Ghanbari University of Essex, UK
	Learning and Performance Evaluation	Prof. Sameer Singh Loughborough University, UK

Copy from http://www.multimediasignalprocessing.com/school/programme

As a researcher I use many tools one I often use is OpenCV and for most things its great but it seems that there port from C to C++ causes problems. The issue I found was with OpticalFlowPyrLK in C++ you will get terrible results it took me a long time to find out what the problem was but the C implementation works fine. So this is more of a note to anyone using OpenCV Lucas Kanade optical flow.

It may not be a suitable sketch for video sketch retrieval but it was the first sketch drawn by someone other than me on "Search Doodle 2.0"

Isn't it beautiful ?

Think back 5 years doesn't seem that long ago but if you think about it people used computers with mouse and keyboards we walked around with small sleak phones with dial pads on there were a few PDA's I remember having an XDA the introductions of the Smart phone age but still based on windows clunky user interface but was the begins even still only a few people had them.

Think back 3 years iOS was released when phones got a kick up the ass a user interface designed around touch also gave a first good app store and many other features that made Microsoft think about there mobile os.

Think back 2 years Android joins the world an open source smart phone OS that introduced a very fast updating phone industry beyond what apple could compete with.

So where are we now, all new phones on the market are smart phones majority of which are touch screen devices, yes we are seeing a resurgence of the slide out keyboard but the touch is at the heart of what they are doing.

So where does sketch come into this, when you pull your phone out your pocket and you hit the google search box to find what you are after on your phone you go back to an old fashion qwerty keyboard that you have to press the keys and hope the gobbledy gook text makes enough sense for your phone to find what you are after. Or you set of few the stylish yet numerous screens of menu's to find what you are after.

This is where the world changes with a massive number of touch aka sketchable interfaces in the hands of the common consumer it is easy to see how an intuitive searching style would help. My work currently relates around videos and how we can search them using sketch the concepts may be simple but the tasks are numerous and great, we may only be able to search for a blob moving right but soon enough we may be able to quickly and easily use a sketch of many other basic objects and add more detail to a search that makes more powerful results possible.

I hope this makes you understand why sketch is important research and keep track of my work and many others in the field.

The main focus of my PhD Is on Sketch and how we can use sketches to retrieve relevant content. Therefore as one of my engineering tasks, I have to create an interface todo this. A while ago I started working on a sketch retrieval search engine although this isn't public yet I thought I would give you a screenshot of the basic interface.

I am currently working on doing some segmentation while testing I generated one of the default windows wall paper into a segmented image and it looks quite cool:

Today I have started an assault on the CVSSP processing servers, they thought they were safe after that last mass of processing done by one of my centre colleagues. I decided to prove them wrong by writing up a few scripts to dump down a load of work on them.

Also played around with some Linux  environmental variables that allows you to switch between platforms and have hostname dependent configurations may come in useful for anyone working with multiple servers.

switch(`hostname`)

      case "workstation**":

            setenv LD_LIBRARY_PATH "/MachineType/opencv:$LD_LIBRARY_PATH"

            breaksw

      case "Sever**Ubuntu":

            setenv LD_LIBRARY_PATH "/MachineType/opencv:$LD_LIBRARY_PATH"

            breaksw

endsw

Note: * is used as wildcard so all servers of that type use the same config

Just a little logon script to change between libraries in my case was in ~/.cshrc