Adjusting photos to correct distortions due to topographic relief, lens distortion, and camera tilt. See Orthorectification
Image classification refers to the process of turning imagery into thematic categories (like the map of Massachusetts, USA below). It is one of the oldest and most common image processing routines. The original image classifier is the photo interpreter; interpreters would draw and label polygons on an aerial photo using a light table. More typical today is classification based on pixel values (Typically referred to as Digital Numbers or DNs) and their categorization using algorithms based on least squares statistics, artificial intelligence, or rule sets. The simplest algorithm classifier is a simple threshold of DNs where pixels with a value above a given threshold belong to one class and pixels with a value below it belong to another. Classifiers, as these algorithms are often referred to, can be categorized into two groups: 1. Supervised, and 2. Unsupervised. Unsupervised classifiers group pixels based on there similarity to each other, forming groups of pixels called clusters. Clusters can then be labeled as classes. Supervised classifiers require a sample of each class' DN characteristics to 'Train' on, pixels in the image are then labeled based on there similarity to the defined samples. The fundamental element of any image classification routine is the classification system, which is just a fancy way of saying a definition of classes. The most common application of image classification is the creation of land cover and land use maps like the one below. So a simple example land use classification scheme would Urban, Suburban, Rural. Image classification can be a applied to any classification system that can be distinguished based on the information contained in the imagery; for example some species level classifications can be extracted using hyper spectral imagery.
Open Source Packages
http://www.digilab.uni-hannover.de/results.html could be used with aerial video footage.
OSSIM Awesome Image Processing
ORFEO Toolbox is an Open Source Remote Sensing Image Processing and Analysis library. It provides ortho-rectification, classification, segmentation, change detection, and many other tools useful for mapping from aerial and satellite imagery.
Leica Geosystems is likely the single largest remote sensing software and sensor provider. they make a remote sensing software called Erdas and photogrammetry suite called Leica Photogrammetry Suite or LPS.
Clark Labs is a non profit research and analysis driven company based out of Clark University that makes the GIS and image processing package Idrisi. Idrisi is a raster GIS and image processing package, so it incorporates many more functions that most. Also, they are non-profit, so its cheaper, and unlike most commercial packages they do not sell extensions (sell you the software piecemeal), one price buys the whole package.
ITT Visual Information Systems makes IDL (Image Data Language) and ENVI, both very powerful image manipulation tools. ENVI is an image processing package and is particularly powerful for hyper spectral image processing. IDL is a programming language that can work alone or in concert with ENVI to complete a whole host of image processing tasks, including custom routines.