ABSTRACT
Convolution neural network (CNN) is a biologically inspired mathematical model of vision. The journey began successfully with the work by David Hubel and Torsten Wiesel who won the 1981 Nobel prize in Physiology or Medicine for this work. The work by Hubel and Weisel was best summarized by the Nobel committee press release ([ppr20]) from 1981. The following paragraph is a reproduction from the press release:
“… the visual cortex's analysis of the coded message from the retina proceeds as if certain cells read the simple letters in the message and compile them into syllables that are subsequently read by other cells, which, in turn, compile the syllables into words, and these are finally read by other cells that compile words into sentences that are sent to the higher centers in the brain, where the visual impression originates and the memory of the image is stored.”
As the quote indicates, the Hubel and Wiesel found that the brain has a series of neurons. The neurons nearest to the retina detect simple shapes such as lines in different orientation. The neurons next to it detects complex shapes like curves. The neurons downstream detects more complex shapes like nose, ear etc.
The understanding of the brain's visual cortex paved way for mathematical modeling of the visual pathway. The first successful work was done by Kunihiko Fukushima ([Fuk80]). He demonstrated a hierarchical model using convolution and downsampling. The convolution al- lowed viewing only a part of the image or video while processing. The downsampling was performed by averaging. Many years later, a different method called "maxpooling" was introduced which is still in use today and will be discussed later in this chapter. The next major breakthrough was the work of Yann Lecun [LBD+89] who introduced a back-propagation approach to learn the parameters of a CNN.
With the availability to large quantity of data, cheaper storage, compute power and software, CNN have become a go to tool for solving image processing and computer vision problems in all areas of science and engineering.
