![Real Time Image Processing : Parallel Implementation On RISC Processor Of Segmentation Chain Based Upon Topological Operators Participants : Ramzi MAHMOUDI, Mohamed AKIL Contacts : {mahmoudr, akilm}@esiee.fr Context & Objectives Approach & Methodology In this project we are studying parallel segmentation’s algorithm based on topological transformation. We focus on two powerful topological operators proposed by ESIEE Engineering - the Computer Science Department : Results Conclusion These are different steps of the new proposed chain. After the dynamic redistribution of the original image (a) we obtain (b). Gradient image (c) is obtained from (b) by the application of Garcia Lorca gradient operator then a double thresholding procedure is applied to get (d). Crest line has firstly been reduced to thin lines (e) by a filtered thinning with =5, then restored (f) by executing the Crest Restoration algorithm until stability. To highlight object of interest a labelling algorithm has been applied to (f) resulting in (g). The new chain provides high segmentation’s quality. The various objects of interest are present in the final image. ,[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],Acquisition Treatment Analysis Displays In this work, we proposed a new chain of image processing, which offers a high quality of segmentation under real time constraints. The new chain is perfectly suited to medical application. Acting directly on grayscale images while preserving their topologies, all topological operators and in particular thinning and crest restoration operators allowed to get a powerful segmentation of 2D brain’s image slices. Medical image processing is a full rising research area : This project makes part of this context : methods of segmentation to isolate one or more anatomical structures in 2D/3D image. Adopted methodology: Constraints: Performance analysis: ,[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],[object Object],Segmentation time for different image resolution Associated cadency (Image 512*512) for each algorithm and architecture enhancement (a) (b) (c) (d) (e) (f) (g) Crest Restoring Operator Thinning Operator Acting directly over Grayscale Image Giving closed and one-pixel thick contours](https://image.slidesharecdn.com/postersegchain-100604133716-phpapp02/85/poster-segmentation-chain-1-320.jpg)
Recommandé
Recommandé
Contenu connexe
Tendances
Tendances (18)
En vedette
Similaire à Poster Segmentation Chain
Similaire à Poster Segmentation Chain (20)
Plus de RMwebsite
Plus de RMwebsite (17)
Dernier
Dernier (20)
Poster Segmentation Chain
- 2. Real Time Image Processing : Parallel Implementation On RISC Processor Of Segmentation Chain Based Upon Topological Operators Abstract Participants : Ramzi MAHMOUDI, Mohamed AKIL Contacts : {mahmoudr, akilm}@esiee.fr In miscellaneous applications of image treatment, thinning and crest restoring present a lot of interests. These algorithms are advised because they act directly on grayscales images while preserving topology. But their strong consummation in term of time and memory remains the major disadvantage in their choice. In this project we propose an efficient hardware implementation of these algorithms on RISC processor allowing to increase the execution time. A chain of segmentation applied to medical imaging will serve as a concrete example to illustrate the improvements brought thanks to the optimization techniques in both algorithm and architectural levels. The particular use of the SSE instruction set relative to the X86_32 processors (PIV 3.06 GHz) will allow a best performance for real time processing : A cadency of 33 images (512*512) per second is assured. Résumé Dans diverses applications de traitement d'image, l’amincissement et la restauration la crête présentent beaucoup d'intérêts. Ces algorithmes sont conseillés parce qu'ils agissent directement sur les images en niveau gris tout en préservant la topologie. Mais leurs forte consommation en terme de temps et de mémoire reste l'inconvénient majeur dans leur choix. Dans ce projet nous proposons une implantation matérielle efficace de ces algorithmes sur processeur RISC permettant d'augmenter le temps d'exécution. Une chaîne de segmentation appliquée à l'imagerie médicale servira d’exemple concret pour illustrer les améliorations apportées grâce aux techniques d'optimisation utilisées aux niveaux algorithmique et d'architectural. L'utilisation du jeu d'instructions SSE relative aux processeurs X86_32 (PIV 3,06 GHz) permettra une meilleure performance pour le traitement en temps réel: une cadence de 33 images (512 * 512) par seconde est assurée.