This is a little knowledge used in choosing the first Project of image major. We need to read an article, then practice the article’s methods and suggest improvements. I will learn relevant knowledge in a more systematic way later, so I just make a simple record of the paper I read for easy review later.

Morphological reconstruction definition

On the basis of morphological gradient image, the gradient image is reconstructed by morphological open and close reconstruction operation, and the details and noise are removed while the important region London is retained. (quoted from blog.csdn.net/lijie45655/…).

Basic knowledge of

The following content is basically a reading of digital image processing (Gonzales) collation, and are binary image cases. This article is also a tidy gonzalez digital image processing, recommend: blog.csdn.net/zizi7/artic…

corrosion

The corrosion of A by B is the set of Z of all the points in A where B is shifted by Z. B here is a structure element. In general, we can use corrosion to remove certain details in the image, which is a shrinking and refining operation.

inflation

The expansion of B over A is the set of Z shifted by Z of all the points where B intersects A. In contrast to corrosion, expansion grows and coarses objects in binary images.

Open operation and close operation

The opening operation of the structural element B on set A is to corrode A with B and then expand the result with B. The closing operation of structural element B on set A is to expand A with B and then corrode the result with B. Personal feeling: The open and close operations passivate the Angle inward and outward respectively.

Morphological reconstruction

Morphological reconstruction involves two images and a structural element. One image is tag image F and one image is template G. Morphological reconstruction algorithm generally relies on constraints and marks to process the image iteratively until convergence, so as to achieve the purpose of automatic processing.

Geodesic expansion and geodesic corrosion

In this case, every expansion and corrosion is constrained by template G. That is, after every expansion and corrosion, it will union or intersect with G. Geodesic expansion and geodesic corrosion of a finite number of images converge due to the limitations of templates after a finite number of iterative steps.

This is a good article: zhuanlan.zhihu.com/p/29401503

Morphological reconstruction of expansion and corrosion is the result of steady convergence of expansion and corrosion.

Re-open operation

The reconstruction opening operation of an image F of size N is defined as the expansion reconstruction after the corrosion from F of size N. In this case, corrosion removes small objects and subsequent expansion restores their shape.

Some other useful basic concepts

Region growth algorithm: imlogm.github. IO Watershed algorithm: studyai.com/article/01b… The explanation of the above two articles is very clear. Note that direct use of the watershed segmentation algorithm results in excessive segmentation due to noise and other local irregularities of the gradient.

Understanding of Adaptive Reconstruction for Seeded Image Segmentation

Adaptive Reconstruction for Seeded Image Segmentation

This paper attempts to find a better morphological reconstruction method, reduce the local minimum value, and make the subsequent watershed algorithm better and more obvious. Traditional multilevel morphological reconstruction uses linear combination to unify the results of multiple reconstruction, but the algorithm proposed in this paper uses nonlinear combination to unify the results of reconstruction. It uses the pointwise Maximum operation for the reconstructed image grids of different units, and the effect is ideal.

subsequent

This article will continue to be updated as the topic progresses. We will first test on binary images and then move to more complex grayscale images. Finally, improvement is proposed.

PS: Part of the content in this article is only the students’ understanding of the book, may not be correct, please check the readers, and welcome to point out the mistakes