|
|
object matching
Mišta, Petr ; Petyovský, Petr (referee) ; Richter, Miloslav (advisor)
Detection of objects based on color is not commonly used method of computer vision. There are many methods thats deals with the detection of significant points, but the color information has been omitted. The goal of this thesis is to design method of the detection significant color image areas and these areas match up with areas detected in another image. I analyzed features of detectors required to identify the reciprocal correspondence of images, defined the color significance concept, described basic color models and their properties, and a design of statistically compiled data - based method was described. Algorithms for the detection of color use color models RGB and HSV. Correspondence of areas detected in different images is performedy Kohonen neural network. The first input vector can teach Kohonen neural network and the second vector is classified by this network. To remove erroneous classifications RANSAC method is used. As a result, the method can be used for basic and rapid determination of correspondence between images, or to speed up commonly used methods for detection of significant points. At the end of the thesis are presented programs, showing functionality and options of design methods. The designed algorithms have been developed in MATLAB.
|
|
|
object matching
Mišta, Petr ; Petyovský, Petr (referee) ; Richter, Miloslav (advisor)
Detection of objects based on color is not commonly used method of computer vision. There are many methods thats deals with the detection of significant points, but the color information has been omitted. The goal of this thesis is to design method of the detection significant color image areas and these areas match up with areas detected in another image. I analyzed features of detectors required to identify the reciprocal correspondence of images, defined the color significance concept, described basic color models and their properties, and a design of statistically compiled data - based method was described. Algorithms for the detection of color use color models RGB and HSV. Correspondence of areas detected in different images is performedy Kohonen neural network. The first input vector can teach Kohonen neural network and the second vector is classified by this network. To remove erroneous classifications RANSAC method is used. As a result, the method can be used for basic and rapid determination of correspondence between images, or to speed up commonly used methods for detection of significant points. At the end of the thesis are presented programs, showing functionality and options of design methods. The designed algorithms have been developed in MATLAB.
|
guest ::
