Lionel Lacassagne
Laboratoire d'Informatique de Paris 6 (LIP6) - Sorbonne University (formely UPMC)
ALSOC team

This site dedicated to High Performance algorithm and embedded systems.
Image processing
  • Connected Component Labeling: the Light Speed Labeling is still the world fastest algorithm
  • Motion Detection algorithms: sigmadelta segmentation
  • Multi-target tracking: agile algorithmes combining mean-shift and covariance matching
more info
Computer Architecture Benchmarking
Benchmarking COTS components with classical image processing algorithms like: Optimized implementations of image processing and media algorithms:
  • Contours and edge detection algorithms: VLIW, Altivec & SSE2
  • Point of Interest: Harris & Achard algorithm for Altivec & SSE2
  • Optical Flow Algorithm: modified Lucas Kanade algorithm for Onera DTIM
  • Wavelet Transform
  • Scan Transform (based on Alain Merigot's idea)
more info
F16: 16-bit floating point: agile FPU for vision-SoC softcore
Developped by ILM, used by NVidia and ATI, coming into Image Processing field...
  • Dicrete Cosinuse Transform (JPEG)
  • Wavelet Transform (JPEG200)
  • Contours and edges detection algorithms
  • Point of Interest
  • Optical Flow: Horn-Shunk
more info
SOG (Special Optimization Group)
  • We design new algorithm and optimize existing ones for all kind of COTS : multi-core SIMD processor, Cell, DSP and GPU
  • (with a very strong expertise on all flavor of SIMD: SSE, AVX, AVX512, Altivec/VMX, Neon, SVE)

  • we also perform some tradeoffs between speed accuracy and power consumption for fit constraints of embedded systems
more info
R&D Industrial Projects
  • Teraops: 1 Teraops of power for embedded plateform
  • OCelle: Tools development for Cell processor
  • SafeAround: to detect and protect
  • LoVE: to avoid pedestrian car crash
more info
PhD position (sujet de th├Ęse), intership (stage), R&D recruitement (CDD), ...
more info
Teaching / Enseignement (french)
Cliquer ici
  • Cours TD et TP