Vernacular--Nonsense image classifier

The following information all is the author's personal understanding, recorded easy to use a simple perspective to understand the complex problems, there is no lack of nonsense and take it for granted, if there is no match with the facts of the place, please make a lot of criticism!

The earliest birth of an image classifier should be a Bayesian classifier:
Why? Because the Bayesian network is actually the various possibilities in series together. Fully consistent with the human brain's reasoning process,
Naive Bayesian classifier: Bayesian networks are actually made up of n possibilities of trees, for example, a red peach in a photograph, a Bayesian algorithm for judging steps such as the following:

A card (90% of the possibility), a poker with Hearts (80%), a card with a letter of Hearts (80%), a poker with a red peach
--a card with hearts with Arabic numerals 4 (20%), a wrong 4 poker in red Peach
A card with spades (20%), a card with a black peach with a letter (80%) and a poker with spades a
A card with a black peach with Arabic numerals 4 (20%), a false spades 4 poker
We look at the above judgment, is not formed a trunk has a forked tree. This is the most important naive Bayesian classifier, the possibility of the formation of a tree, so called "simple (tree)", then in the main areas of Bayesian thinking, there are some people to do some optimization, nothing more than to add a number of parameters such as the threshold, the tree into a diagram, so that the possibility of nodes can flow between the branches. The aim is to make the classification more accurate

Although very easy to understand, the naive Bayesian classifier cannot accurately judge and classify complex image data. The digital image information is huge, naive Bayesian classifier can only be classified for the image with obvious feature or one-dimensional space can be divided.

With the development of computer computing ability, it is possible to make large-scale classification operation. So the neural network classifier was born. A neural network is a way of simulating how people's brains work. A large number of classifier nodes are used to simulate the "branch" of the human brain, and the operation is carried out.

Each "axon" classifier node can be a "Bayesian classifier"
Artificial neural Network: A collection of n multiple classifiers similar to the "Bayesian algorithm". Together to judge.

watermark/2/text/ahr0cdovl2jsb2cuy3nkbi5uzxqvemvyz3nrag==/font/5a6l5l2t/fontsize/400/fill/i0jbqkfcma==/ Dissolve/70/gravity/southeast "> Note: Every blue dot is considered a Bayesian classifier, which makes it easy to understand

Well, see here. Perhaps it is clear why so many people are now studying neural networks? As a result of this system architecture, the space and potential to play is enormous
But this kind of architecture is not a cure, really determines the image classification effect. is also the algorithm in each detailed node, and the participation of a large number of nodes. Just a kind of "human sea Tactics" (Machine Sea)

So people start thinking about how to make complex and difficult to classify image data. became available, so the support vector machine classifier was born.
Support Vector machine classifier: In a nutshell, the non-divided data in the low-dimensional space is transformed into a high-dimensional space, which can be divided into

Support Vector machine The name is too academic and daunting, in fact it's not that complicated.

Example. is still the above-mentioned red peach a poker. What is the working process of support vector machines?

Digital Image Two-dimensional data (x-axis data, y-axis data)--(whether or not)--three-dimensional data (x-axis grayscale, y-axis grayscale, overall color values)--(Can be divided)- > Four-dimensional data (x-axis grayscale, y-axis grayscale, x-axis color, y-axis color) ...

See here, you may be clear, support vector machine is the original mixed irregular data according to a certain rules to filter out. Look for the filtered data until you find a pattern.

so. The quality of support vector machine classification depends on the setting of the filtering rule, which is called kernel function.

So support vector machine really from the meaning of the digital image of the content of learning, so it is now a hot research!

wrote here today, hoping to help you understand these concepts, the traditional material always write a lot of mathematical formulas, in fact, very many are not necessary, with the formula to express the concept, is the mathematics of Classical Chinese, a bit of the meaning of Kazakhstan.

Vernacular--nonsense image classifier