1. Scientific Writing Abstract Writing

2. Why ? Most important part of the paper Number of Readers ! Make people read your work. Sell your work. Make your work searchable ?

3. Abstract Writing What ? Motivation – Why do we care ? Problem statement – What’s the problem ? Approach – How did you solve it ? Results – What’s the answer ? Conclusions – What are the implications ?

4. Abstract Writing Types of abstracts Descriptive Abstract Motivation, Problem No approach, no results, no conclusion < 100 words Informative Abstract Motivation, Problem Include approach, results, conclusion

5. Abstract Writing How? Once you’re done with the writing Read in one batch Lean back – and wait (glass of red wine) Then write abstract in one batch without looking back to the text Take another sip from your glass of red wine Then read abstract and re-iterate until convergence (more red wine)

6. Abstract Writing What else ? Keep it short (< 200 words) No bibliographic references No mathematical formulas No explanations – just facts Use the right key words (search engines) Don’t’ copy text - reformulate

7. Abstract Writing Abstract – Summary for your Thesis Not more than 1 page ! Place it before table contents. Double-check – better triple-check for typos and spelling errors.

8. Abstract 1 The paper reviews methods for computing the intersection of two subspaces. The problem arises when two measurement processes produce alternative predictions for an observed event and where these predictions need to be combined or fused. The paper reviews the notion of a parallel sum of projection operators, as introduced by Anderson and Dun for the computation of the projector onto the intersection of two subspaces. For computational reasons, the parallel sum is determined using a Schur complement scheme. This scheme allows for an interpretation in terms of multi-ports, where multi- ports are engineering models originating in circuit theory. As an alternative to computing the parallel sum, the paper reviews an iterative approach to compute the intersection, which is based on a result published by Nakano and Halmos.

9. Abstract 2 In this paper a new method of detecting and tracking a human person in three dimensional space using audio and video data is proposed. A simple tracking system with two microphones and stereo vision is introduced. The audio information is resulting from the Generalized Cross Correlation (GCC) algorithm, and the video information is extracted by the Continuously Adaptive Mean shift (CAMshift) method. The localization estimates delivered by these two systems are then combined using a novel Particle Swarm Optimization (PSO) fusion technique. In our approach the particles move in the 3D space and iteratively evaluate their current position with regard to the localization estimates of the audio and video module. This facilitates the direct determination of the object’s three dimensional position. Compared to existing methods, this novel technique achieves faster tracking performance while being independent of any kind of model, statistic, or assumption.

11. Abstract 3 With the advances in modern cameras, the size of digital images is increasing rapidly. This fact poses a limitation on stereo matching algorithms since the operation on images with millions of pixels requires a huge amount of resources. The problem can be solved by processing the images at low resolution and then upsampling the result. This solution, however, faces the limitations of the image reconstruction methods. In this paper, we propose a technique based on the theory of compressed sensing to reconstruct a higher resolution disparity map from its lower version. The key issue is to assume that the disparity values of the lower resolution image as random sparse measurements of the high resolution disparity map. We then formulate a constrained minimization scheme to recover the latter from the measurements. Tested on the Middlebury ground truth data set, the algorithm is able to retain a good quality. Using as low as 10% of the pixels, the reconstructed disparity maps remained at the first rank in the table. Compared to other methods, the scheme leads to an improvement in the quality.

12. Abstract 4 We propose a Bayesian framework for representing and recognizing local image motion in terms of two primitive models: translation and motion discontinuity. Motion discontinuities are represented using a non-linear generative model that explicitly encodes the orientation of the boundary, the velocities on either side, the motion of the occluding edge over time, and the appearance/disappearance of pixels at the boundary. We represent the posterior distribution over the model parameters given the image data using discrete samples. This distribution is propagated over time using the Condensation algorithm. To efficiently represent such a high-dimensional space we initialize samples using the responses of a low-level motion discontinuity detector.

13. Scientific Writing Abstract Writing

15. Introductory Chapter First chapter of your thesis 1.Introduction – Motivation – Context 2.State of the art – Previous work 3.Problem statement – Unsolved issues 4.Overview and main results