Image Recognition Methods

OpenCV

Breaking out Features

Local Binary Patterns (LBP):

OCR Optical Character Recognition

Image correlation: Optical Mice

Browser Based Image Processing: Subtraction, Area Sum

Methods for detecting Optical Flow +

Synthetic Data: To collect large sample sets for character recognition, you can synthesize "fake" data (Artificial Data) by randomly pasting letters from a randomly selected font onto a randomly selected background with (perhaps) random distortions and rotations. The major advantage of doing this is that you don't have to "label" the data (know what character(s) are represented) because you made it from the label in the first place. It can also help in trouble shooting complex systems, by replacing a stage in a complex system with a "perfect" stage to see how much that effects the end product. E.g. if you are wondering how much background removal helps your OCR application, generate input images with no background.

Also:

• Neural Networks especially Convolutional Neural Networks (CNN)
• Logistic Regression
• Soon Lee's MassMind page Image classification, labeling, and pattern recognition.
• MassMind Newsletter October 2003: Ranging via Camera and Laser
• MassMind Newsletter August 2003: Easy Vision Systems

See also:

• http://hackaday.com/2016/04/02/insanely-quick-3d-tracking-with-1-camera/ Very fast, Single camera, 3D tracking and scanning. Open source code supports ROS.+
• http://www.robots.ox.ac.uk/~misard/condensation.html The Condensation algorithm (Conditional Density Propagation) algorithm tracks moving objects through substatial clutter.
• http://neil.fraser.name/software/recog/
• http://www.bitquill.net/trac/wiki/Android/OCR (see the blog post) Given a greyscale image of dimisions m x n, the lighting will often change from one part of the image to another. To pick out details consistantly, you need to set the threshold individually for each pixel (i,j) as the mean intensity of pixels in a w x w neighborhood around the pixel. On each row, you might do
  for i = 0 to N-1: s = 0; for k = max(i-r,0) to min(i+r,N-1): s += a[k];
  where w=2r+1. (r is half the width of the window). But it turns out that this is faster:
  Initialize s = sum(a[0]..a[r]); for i = 1 to N-1: if i > r: s -= a[i-r-1]; if i < N-r: s += a[i+r]
  at the end of each, s is the threshold for pixel a[i].
• http://www.cs.colostate.edu/~draper/papers/draper_cviu05.pdf B. Draper and A. Lionelle. “Evaluation of Selective Attention under Similarity Transformations,” Vision and Image Understanding, 100:152-171, 2005
• http://www.cs.colostate.edu/~draper/research/software.php
•