@article{meshgi2015phd,
  title={Brain Inspired Face Detection},
  author={Kourosh, Meshgi},
  year={2015},
  publisher={Amir Kabir University of Technology},
  abstract = {Encoding the building blocks of a natural image, such as edges, textures and shapes is one of the main functions of a visual system which is the underlying procedure of visual high level tasks such as face detection and natural image understanding. An abstract representation which is generalized over specific samples of input images is the desired output of this encoding. It’s now understood that primary visual cortex employs a combination of features extracted from visual data but there is not a prominent theory about forming invariant representation in human brain. Higher-order visual neurons are commonly assumed to use the statistical variations that characterized local image regions. In the proposed model, a search about the most consistent distribution to the input images is conducted and represented using activities of the neurons. This model is feed with natural scenes and human face images and obtains generalization over this training set by learning a compact set of dictionary elements typically found in natural images. The characteristics of model neurons resemble cortical cells, and provide an explanation for nonlinear properties of these cells functionally. Being inspired from coding strategies underlying the processing of primary visual cortex and higher visual areas, the results representation is able to correctly cluster images with similar distribution with is a part of concept formation. Finally a face detection algorithm is proposed which use this coding strategy and other mechanisms inspired from brain neural network. The proposed model is invariant against changes in illumination and facial expression changes and is capable of detection and localizing of faces appeared in a still image. The robustness of the algorithm in cluttered backgrounds and its ability to find multiple frontal faces including occluded ones are another features of the proposed model as demonstrated by the experiments.}
}