@TechReport{Sochman-TR-2004-04,
  IS = { zkontrolovano 30 Mar 2004 },
  UPDATE  = { 2004-03-23 },
  author =   {{\v S}ochman, Jan},
  title =   {{AdaBoost} for Fast Face Detection -- {PhD} Thesis
                  Proposal},
  institution =   {Center for Machine Perception, K13133 FEE Czech
                  Technical University},
  address =   {Prague, Czech Republic},
  year =   {2004},
  month =   {March},
  type =   {Research Report},
  number =   {{CTU--CMP--2004--04}},
  issn =   {1213-2365},
  pages =   {34},
  figures =   {5},
  authorship =   {100},
  psurl =   {PDF},
  project =   {GACR 102/02/1539, CTU 0307313},
  annote =   {The main objective of our research is to study the
                  two class example-based learning problems with
                  different measurement costs for the classes, where
                  time of the classifier evaluation is limited or
                  where the evaluation time is the optimisation
                  parameter. The face detection problem is used as an
                  example of such problem, nevertheless the results
                  should be applicable to the other similar
                  problems. For the classifier training the AdaBoost
                  algorithm is used. Two improvements of the
                  state-of-the-art algorithms are proposed in this
                  thesis proposal. The totally corrective algorithm
                  with coefficient updates improves the evaluation
                  time of a single classifier trained by the AdaBoost
                  algorithm by minimising the training error upper
                  bound more aggressively and producing shorter
                  classifiers. The second proposed enhancement, using
                  the previous-stage knowledge during the training of
                  a cascaded classifier shorten the classifier average
                  evaluation time by a partial sequentialising of the
                  cascade building training and speeds up the training
                  itself. Since many problems in computer vision are
                  of the studied form a proper formulation of these
                  problems and development of the corresponding theory
                  will lead to better understanding of these problems
                  and to more efficient solutions.},
  keywords =   {pattern recognition, computer vision, AdaBoost, face
                  detection},
  comment =   {minimum},
}