GRASS GSoC 2016 Additional Image Segmentation Algorithms for i.segment
|Student Name:||Bo Yang|
|Organization:||OSGeo - Open Source Geospatial Foundation|
|Mentors:||Moritz Lennert, Markus Neteler, Markus Metz|
|Title:||Additional segmentation algorithms for i.segment|
|Repository:||GRASS 7, browse at: i.segment|
GRASS GIS has the i.segment which provides the possibility to segment an image into objects. This is a basic step in object-based image analysis (OBIA). Currently, the module only provides one segmentation algorithm: region-growing. The code of i.segment was structured in a way that allows addition of other algorithms. The core of proposed GSoC project would thus be to add a series of these algorithms. It would be more useful and comprehensive to add more segment methods to the i.segment module, such as mean-shift and watershed, which could be used in more types of satellite image processing. Special care should be taken for the whole project to code as efficiently as possible, i.e. to make the code run in reasonable time, even for very large images.
Idea for this project was suggested by Moritz at GDAL SoC Ideas.
- Image segmentation or object recognition is the process of grouping similar pixels into unique objects. Segmentation of remote sensing images is a challenging task. A myriad of different methods have been proposed and implemented in recent years. In spite of the huge effort invested in this problem, there is no single approach that can generally solve the problem of segmentation for the large variety of image modalities existing today. The most effective segmentation algorithms are obtained by carefully customizing combinations of components. The parameters of these components are tuned for the characteristics of the image modality used as input and the features of the objects to be segmented. 
- In the GRASS i.segment module currently only region growing and merging algorithm is implemented. Each object found during the segmentation process is given a unique ID and is a collection of contiguous pixels meeting some criteria. Note the contrast with image classification where all pixels similar to each other are assigned to the same class and do not need to be contiguous. The image segmentation results can be useful on their own, or used as a preprocessing step for image classification. The segmentation preprocessing step can reduce noise and speed up the classification. 
- 1. Region growing and merging (available in i.segment module )
This segmentation algorithm sequentially examines all current segments in the raster map. The similarity between the current segment and each of its neighbors is calculated according to the given distance formula. The basic approach of a region growing algorithm is to start from a seed region (typically one or more pixels) that are considered to be inside the object to be segmented. The pixels neighboring this region are evaluated to determine if they should also be considered part of the object. If so, they are merge to the region and the process continues as long as new pixels are added to the region .
- 2. Mean-shift (plan to be implemented during GSoC 2016)
The mean shift segmentation is a local homogenization technique that is very useful for damping shading or tonality differences in localized objects. For the algorithm implementation of this case, basically the algorithm replaces each pixel with the mean of the pixels in a range-r neighborhood and whose value is within a distance d. The Mean Shift takes usually 3 inputs: 1) A distance function for measuring distances between pixels. Usually the Euclidean distance, but any other well defined distance function could be used. The Manhattan Distance is another useful choice sometimes. 2) A radius. All pixels within this radius (measured according the above distance) will be accounted for the calculation. 3) A value difference. From all pixels inside radius r, we will take only those whose values are within this difference for calculating the mean .
- 3. Watershed (plan to be implemented during GSoC 2016)
Watershed segmentation classifies pixels into regions using gradient descent on image features and analysis of weak points along region boundaries. Imagine water raining onto a landscape topology and flowing with gravity to collect in low basins. The size of those basins will grow with increasing amounts of precipitation until they spill into one another, causing small basins to merge together into larger basins. Catchment basins are formed by using local geometric structure to associate points in the image domain with local extrema in some feature measurement such as curvature or gradient magnitude. The watersheds technique is also more flexible in that it does not produce a single image segmentation, but rather a hierarchy of segmentations from which a single region or set of regions can be extracted a-priori, using a threshold, or interactively, with the help of a graphical user interface.
- Preparation: Discuss with Mentors. Gather ideas from the community. Feature requests, image segmentation literature, and any other ideas and suggestions.
- 16 – 21 May week 0: Setup coding environmental, get familiar with programming manual, test through existing code.
- 23 -- 28 May week 1: Start coding, develop pseudo code to outline the work
- 30 May -- 4 June week 2: implement mean-shift image segmentation algorithm
- 6 -- 11 June week 3: Validation mean-shift algorithm
- 13 -- 18 June week 4: Debugging mean-shift algorithm
- 20 -- 26 June Week 5: Caching week, further refine and validate the code
- 27 June Mid-term evaluation: Evaluate the existing program, determine the plan for the remaining 3-4 weeks.
- 28 June -- 2 July Week 6: based on the evaluation of the mid-term, test and ensure a solid existing program.
- 4 -- 9 July Week 7: Implement watershed image segmentation algorithm
- 11 -- 16 July Week 8: Validation and debugging watershed algorithm
- 18 -- 23 July: Further refine tests and documentation for the whole project.
- 25 July – 13 August Week 9-11: Improving the main algorithm, if time permits, adding shape as an additional merge criteria,
- 15 August - 23 August Final week: Tidy code, write tests, improve documentation and submit code sample.
Implement more image segmentation methods to extend the available i.segment for image processing in GRASS. As the general logistics of the i.segment module is in place, adding mean-shift and watershed segmentation algorithms should be possible. The core of the GSoC project thus is to add a series of these algorithms. In addition, the current implementation only uses distance within the multidimensional space of all input bands as the criteria whether to merge segments or not. Adding shape as an additional merge criteria will be helpful. If time permits, this feature of additional shape criteria will be implemented.
- Get an OSGeo user id : this will give you access to trac, including the wiki part of trac
- Set up a complete build environment allowing you to check out different versions of GRASS, compile and run them 
- Read the programming manual , notably the chapters General (aka GIS Library), Imagery, Raster, Segment
- Read submitting (coding standard) rules , especially the C-rules.
- Register on the OSGeo wiki (making yourself automatically a member of OSgeo)
- More will be added...
Possible difficulties and solution
As it usually happens with software development, I might come across problems during implementation of the above mentioned ideas which may take longer than I expected, in such a case, I have kept two weeks (week 5 and week 9) as caching weeks which are intended to let me catch up on unimplemented parts of the project. I will be working on this project full time as of now. I plan to keep in touch with my mentor as much as I can and will be reporting to him every time I am done with finishing things on my to-do list. My college semester ends on May 5. New semester starts at the middle of August and hence only about one weeks of the GSoC project would overlap with college work, this will not be a problem as there isn’t much work in the first few weeks of college. After that, I have no other obligations whatsoever and am completely free to work on the project. If my project need more work or maintenance after the summer ends, I would love to stick around and help. I actually have an own idea and want to implement and add to the GRASS lib after the training through GSoC. Afterwards I will look at other projects and see what I’m interested in.
My name is Bo Yang, I have finished Master degree in Geography at University of Cincinnati, now I am in the second year of Ph.D. study of Geography at UC. I also have a bachelor degree in Mathematics and MS in Computer Science. I like Python language, but I have also used other languages (C language, PHP), and have utilized QGIS and GRASS a lot in my study and research project.
- Email: firstname.lastname@example.org
- Address: 477 Riddle Rd, Cincinnati, OH 45220
-  OTB software guide: https://www.orfeo-toolbox.org/packages/OTBSoftwareGuide.pdf
-  GRASS Manuals: https://grass.osgeo.org/grass71/manuals/i.segment.html
-  Comaniciu, D., & Meer, P. (2002). Mean shift: a robust approach toward feature space analysis. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(5), 1–37.
-  http://stackoverflow.com/questions/4831813/image-segmentation-using-mean-shift-explained