readme

基于贝叶斯理论的指纹识别算法及学习套件, 使用贝叶斯概率论实现对指纹识别,特征码提取,特征对数获取的功能

Student: Scott Sanner
Email:   [e-mail]
Course:  CS223B, Winter
Final Project: Rowley-Beluja-Kanade Face Detector

System Requirements
===================
Matlab 5.x with the Image Processing and Neural Net
toolboxes.

Unpacking the files
===================
Create a directory for the project and untar the tarfile
using the command 'tar -xvf fd.tar'.  This should create
a number of m-files in the current directory and a
subdirectory of images used to train the neural net.

Running the Face Detector
=========================
To build a trained, face-detection neural net, simply
run the 'facetrain' script.  The two global variables needed
from this script are 'NET' which is the trained neural net,
and 'MASK' which is the mask used to define the area inside
a rectangle which will be tested for facehood.

To run the face-detector on an image using the 'facescan'
function, simply pass 'NET', 'MASK', a double array of the 
grayscale image, and a few parameters governing the 
detection threshold and image scanning characteristics.  
Check the 'facescan.m' file for more information on good
values for the paramters.

File Listing
============

M-Files (Main Files)
--------------------
facetrain.m     - The main neural net training script
                  which loads all required training files,
                  builds the needed data structures,
                  and trains and tests the neural net

facescan.m      - The main image face-scanning function.
                  Simply pass this function the neural
                  net, image, mask, and a few parameters
                  governing the detection threshold and
                  scanning characteristics.  See the
                  contents of this file for for good
                  default values for the parameters.

M-Files (Neural Net Utilities)
------------------------------
createnn.m       - Creates a neural net given the input,
                   hidden, and output unit characteristics.
simnn.m          - Simple formats the data for presentation
                   to the neural net.
trainnn.m        - Trains a neural net given labeled data
                   and a percentage to use for the 
                   validation set (which it constructs).
classifynn.m     - Normalizes an image and returns the
                   classification value from the neural net.

M-Files (Image Utilities)
-------------------------
buildimvector.m  - Builds an image vector from a rectangular
                   image array.  Used to convert data so
                   that it can be used by the neural net
                   for training.
buildresvector.m - Builds a result vector to match the
                   image vector.  
buildmask.m      - Builds a rectangulary binary mask array
                   for face images.
normalize.m      - Normalizes an image by subtracting a
                   linear lighting plane and rescaling the
                   grayscale distribution histogram.
augmentlr.m      - Augments an image set with the left-right
                   flipped versions of the images.
augmentud.m      - Augments an image set with the upside-
                   down versions of the images.

M-Files (Image Loading/Display)
-------------------------------
loadimages.m     - Loads a set of images according to the
                   given pattern set.
showimages.m     - Subplots a set of images in an image
                   array.

Image Data (Using wildcards '#')
--------------------------------
scaled/n##-x.PNG - Non-face files used for training
scaled/s##-n.PNG - Face data of normal pose for s##
scaled/s##-c.PNG - Face data of center lighting pose for s##
scaled/s##-l.PNG - Face data of left lighting pose for s##
scaled/s##-r.PNG - Face data of right lighting pose for s##
scaled/s##-g.PNG - Face data of pose with glasses for s##