Research Interests
  • Image Processing and Computer Vision
    • Human Motion Recognition
    • Local Features (SIFT, SURF, etc.)
    • Face Recognition and Tracking
    • Facial Feature Extraction(especially eye and eyebrow)
    • Super Resolution(especially from multiple frames/images)
    • Edge and Corner Detection (Canny, Wavelet, CLA, etc.)
  • Artificial Intelligence and Robotic
    • Human Computer Interaction (especially base on face elements tracking)
    • Swarm Intelligence and optimization (PSO, Ants, etc.)
    • Optimization (Simulated Annealing, GA, etc.)
    • Multi Agent Systems
    • Game theory
Research Projects

Complex Network of Images, A novel Image Search system
Contributor(s): Mehdy Bohlool
Files: Presentation and report files will apear soon
Abstract: Searching a large database of images based on the keyword or content is a very active research topic. Time complexity of these search engines as well as quality and relativeness of results are hugely depend on the labels of the images in the database as well as how they are connected. Any set of images can be considered as a network of images connected based on some sort of similarities between them. At the same time, recent advances in complex networks revealed interesting properties in social networks such as social capitals and epidemics dynamic. This study will investigate these properties on the image network and utilized them to find better candidates for labeling as well as spread labels more efficiently through the network. Our experiments shows that the new image search system returns more relevant images with less labeling.

Human Motion Recognition using Dimensionality Reduction
Contributor(s): Mehdy Bohlool
Files: Presentation (PDF) , Presentation and Animations (ZIP) , Paper (PDF)
Abstract: In this project, the effect of different classifiers and dimensionality reduction methods is studied on the problem of human motion recognition and classification. The high dimensionality nature of human motion data prevent them to be recognized in the original space, thus they should be mapped to a lower dimensionality space for better result or at least better performance. Different dimensionality reduction methods such as LLE, Isomap, MDS is applied to a standard dataset and the result of recognition using kNN and QDA classifiers with different parameters are compared. To compare the result of the classifier and find a good classifier for this domain and dataset, a standard statistic comparison method called Fisher exact sign test is used.

Computer Vision Syndrome prevention using real-time accurate Blink Detection
Contributor(s): Mehdy Bohlool
Files: Sample Video , Paper (PDF)
Abstract: Computer Vision Syndrome (CVS) is a common problem among computer users. Staring at an object from two feet for a long time will make the eye dry and cause eye strain and fatigue. Most of the computer users suffer from headaches, loss of focus, burning eyes, and blurred vision during their long time work in front of a computer monitor. In this paper a computer vision software solution (CVSS) to CVS is proposed to decrease these symptoms and prevent dry eye effect. CVSS traces user eyes in real time, through a simple low resolution web cam, that exists in most modern computers and laptops. No special hardware or configuration is required to use this software. The system counts eye blinks and logs their durations and warns user to blink regularly thus avoiding dry eye. It also tracks user’s blink frequency and other parameters to warn the user about drowsiness and exhaustion of the eye or collect this information for oculists. Experiments show that the system can detect and analyze blinks in real time with low false- positives and very low false-negatives and it succeeds in increasing users blink rate through various notification methods.

Cost-efficient Automated Visual Inspection System for small manufacturing industries based on SIFT
Contributor(s): Mehdy Bohlool, Soroosh Rahimi Taghanaki
Files: Paper (PDF), IEEEXplore Link
Abstract: This paper presents a cost efficient Automated Visual Inspection (AVI) system for small industries’ quality control system. The complex hardware and software make current AVI systems too expensive to afford for small-size manufacturing industries. Proposed approach to AVI systems is based on an ordinary PC with a medium resolution camera without any other extra hardware. The Scale Invariant Feature Transform (SIFT) is used to acquire good accuracy and make it applicable for different situations with different sample sizes, positions, and illuminations. Proposed method can detect three different defect types as well as locating and measuring defect percentage for more specialized utilization. To evaluate the performance of this system different samples with different sizes, shapes, and complexities are used and the results show that proposed system is highly applicable to different applications and is Invariant to noise, illumination changes, rotation, and transformation.

A novel Edge Detection Method based on Dissipative Cellular Learning Automata
Contributor(s): Mehdy Bohlool, Mohammad Reza Meybodi
Files: Translated Paper (PDF), Full Paper (PDF, Persian)
Abstract: Cellular Learning Automata (CLA) is a model for systems consisting of simple elements. These simple elements improve their actions based on theirs neighbors behavior and their last experiences. Nevertheless, they can expose complex behavior based on their interactions. Open and asynchronous Cellular Learning Automata are similar to CLA except that the process of cell updating executes asynchronously and each cell’s behavior depends not only on its neighbor's actions but also on extra global factors. In this paper, a new method for edge detection based on open and asynchronous CLA is proposed and compared with classic Canny Edge Detection method. The proposed method is less sensitive to noise and finds more continuous edges than the Canny Operator. Experiments show that the proposed method has good performance and is less sensitive to noise and texture.

Etching Process Simulation using Dissipative Cellular Automata
Contributor(s): Mehdy Bohlool, Mohammad Reza Meybodi
Files: Translated Short Paper (PDF), Full Paper (PDF, Persian), Simulation Video
Abstract: Etching Process is one of fundamental processes in manufacturing Micro Electronic Mechanical Systems (MEMS) that usually simulated by Cellular Automata. The main problem of existing simulation using CA is that All cells updates synchronously that is not confirm with real world processes, and there is more than one real crystal cell in one single simulation cell that will reduce simulation accuracy. In this paper a simulation model based on Dissipative Cellular Automata proposed and well implemented. The result of this simulating compared with Scattering Electro Microscope (SEM) pictures and shows good matching.

©, All right reserved

Subpages (1): Publications