Identify a Specified Fish Species by the Co-occurrence Matrix and AdaBoost - Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing - page 98

Fig. 2 Fish detection result with the learned feature file

3.1 Extraction of Texture Feature

Texture plays a very important role in visual information processing. It is thought
that there are two texture level[20], structural texture and statistical texture[21, 22].
Considering from the standpoint of texture image analysis, more stronger the struc-
ture is, more easy for analysis. However , those nature texture is almost exists in
statistical manner. And there is no way to analyze the texture in unique over the
difference texture.
Broadly speaking the extraction of texture features of conventional can be dev-
ided in to:

(1) Extraction of statistical features.
(2) Analysis by local geometric feature.
(3) Analysis by fitting model..
(4) Structural analysis

Most of the analysis method is defined mathematically, which are not taken into
account of correspondence between the visual psychology, but it is a quite effective
way.
Texture features so strongly affected by the illumination of the image shooting, as
a pretreatment, normalize the density of the image before extracting texture features
is good in many cases.
As a typical calculation method of texture statistical characteristics, there are
density histogram, co-occurrence matrix, the difference statistic, run length matrix,
the power spectrum. Texture analysis is referred to as a N th order statistical value
relate to the combination of the concentration of point d at the certain position rela-
tionship.
Statistics introduced from the density histogram is 1st order statistics, statistics
derived from co-occurrence matrix, difference statistics and power spectrum are the