Research Projects
Data set
Source Code
Media Coverage

Research fields

Computer Vision
Signal Processing
Machine learning
Computational Neuroscience

Research Interests

Sparse approximation
Inverse problems
Real-time vision
Bio-inspired vision
Large-scale vision
Big Visual data

Here are my on-going projects in:

I- Human behavior analysis
II- Low-level image descriptors
III- Computer graphics: Video Processing
IV- Large-scale signal processing

    I - Human behavior analysis

Foreground silhouette extraction robust to sudden changes

foreground silhouettes Human traffic analysis in a network of fixed cameras depends on the quality of the extracted foreground silhouettes. Monocular background subtraction algorithms often model the variation in pixels intensity hence are sensitive to lighting conditions. We have developped a framework robust to sudden changes of illumination such as a spot light effects in exhibitions or projected videos in the background.

People detection in high-density crowds

people detection High density crowds are best analyzed with a dense network of fixed cameras with overlapping field-of-views. We have developped a dictionnary-based framework with sparsity promoting priors to locate people in real-time with a network of calibrated cameras.

People tracking in extreme conditions

People tracking Most multi-view systems assume a well structured environment to detect and track objects across cameras. The latters need to be fixed and calibrated. We have designed a cascade of image descriptors to track objects across a sparse network of un-calibrated cameras without any overlapping field-of-view.

Human flow analysis

behavior analysis

To best understand human traffic, thousands or even millions of extracted trajectories need to be summarized and analyzed. We present a sparse representation to efficiently identify most important trajectories and rank them.

(coming soon)


    II- Low-level image descriptors


FREAK: Fast Retina Keypoint

freak The last decade featured an arms-race towards faster and more robust keypoints and association algorithms: SIFT, SURF, and more recently BRISK to name a few. We have developped a keypoint descriptor inspired by the human visual system and more precisely the retina, coined Fast Retina Keypoint (FREAK).FREAKs are in general faster to compute with lower memory load and also more robust than SIFT, SURF or BRISK.

Image-based retrieval: search within images

visual search There is a tremendous amount of public images on the internet to retrieve. The amount of private images is also exponentially increasing due to the deployement of high quality sensors on mobile phones. We have developped a framework to cluster and retrieve images efficiently using FREAK descriptors.

Face & eye detection

Face eye detection There is a new trend to embed cameras on new devices such as TV or portable gaming devices to offer new experiences. One domain of application is to propose fake 3D, or glass free 3D screens by automatically detecting the positions of the human eyes. However, existing algorithms are to expensive to run on low-cost hardwares. We have designed a low-computing face and eye detector to suit the limited requirements in chip design.


    III- Computer graphics: Video processing

Video Re-targetting

image retargetting The continuous development of new display devices (e.g. mobile phones, notebooks) induces a constantly growing consumption of media content. The devices may have different resolutions and aspect ratios and a resizing of the image might be needed to enable its display on a particular device. We have developped a new content-aware image resizing scheme, Stream Carving, which is based on the well-known seam carving method. Our retargeting algorithm is related to human perception by exploiting an adaptive importance map that merges several features like gradient magnitude, saliency, face, edge and straight line detection.

Video Stabilization

video stabilization

The deployement of HD video cameras on smart phones motivates the need to efficiently stabilize videos captured on the fly. An optimized algorithm has been developped to perform as good as state-of-the-art while running real-time.

(coming soon)


    IV- Large-scale signal processing


Crowd-sourced adaptive content-based media recommendation

Graph matching Content-based media recommendation requires advanced signal processing algorithms to analyze and compare the medias. The computation time increases exponentially with the number of available medias. Typically, there are millions of media songs that need to be analyzed to find similar music. We have developped a framework to address the presented challenge to identify similar medias content based on signal analysis given a tremendous amount of data.




© alahi {at}
updated: January 2015

Follow me on

Alahi on ScholarAlahi on LinkedinAlahi on Twitter


Schools Studied


Ecole Polytechnique Fédérale de Lausanne

Labs Worked

Vision lab at Stanford



Companies Worked