Study of image segmentation methods

The article describes the study of image segmentation methods in various examples. The purpose of the study is to detect the advantages and disadvantages of some known methods.

Methods that will be discussed in this article:

1. Method of growing regions Here you can read about the method of growing regions ;
2. The method of the watershed And here in detail about the method of the watershed, also with the code ;
3. Method of normal cuts Read more to read here .

Study of segmentation methods on model images

A study of segmentation methods was initially conducted on image models. Nine types of images were used as models.

Method name	results
Method of growing regions
Watershed method
Normal cuts method

The results of the study showed:

• The method of growing regions localizes texture defects, both sharply different from the background, and formed by turning and changing the brightness of the texture;
• The method of growing regions to a varying degree localizes defects at different angles of rotation of the texture;
• The considered method of segmentation of the watershed in its original form does not provide localization of textural defects;
• The method of normal cuts well localizes the presence of a texture different from the background, but does not highlight the change in brightness and rotation of the texture.

Study of segmentation methods on object images

To study segmentation methods, a database of images of various objects was prepared. The resulting images were segmented using various methods, the result of which is shown in the figures in the table.

Source image	Method of growing regions	Normal cuts method	Watershed method

Results:

• The method of growing regions does not provide for the localization of segments in the images of an object;
• The considered methods of the watershed and normal sections in the original form do not provide localization of the objects presented;
• The method of normal cuts provides localization of objects in the images of objects.

results

The results of the study:

• The method of growing regions does not provide for localization of segments both on model images and on object images, and also provides localization of road infrastructure elements.
• The considered methods of the watershed and normal sections in the original form do not fully provide for the localization of the objects presented.
• The method of normal cuts provides localization of objects both on model images and on images of objects, and also provides localization of road transport infrastructure elements.
• The method of growing regions and the method of normal cuts can be recommended for use in automated visual inspection systems.