Abstract, Introduction, & Results and Discussion
In various fields, quick and accurate particle size distribution analysis is desired. Currently, many distribution analyses and created manually, which leaves room for inconsistencies and can be fairly time consuming. An ImageJ plug-in has created a way to use computer (vision-based) image processing as an alternative or replacement method for measurement, identification, and size distribution analysis.
Though ImageJ comes with a built-in option for analyzing particles, this option presents multiple problems; including not providing dimensions of practical interest (i.e. the length and width of the particles), large deviations from the actual dimensions when using the available dimensions from the bounding rectangle of the particle, and deviations influenced by the particle shapes. Thus, it was concluded that fitted ellipse dimensions be used as opposed to the rectangle dimensions. But, these dimensions still showed some deviation, such that they still need to be corrected for accuracy. The research work proposed deals with developing an ImageJ plug-in that is able to analyze particle size distribution by applying correction factors after determining the particle shapes and then accurately give the dimensions of the particles through corrected fitted ellipse dimensions.
The Results and Discussion section discusses several aspects of the ImageJ plug-in that were tested. One aspect is how using the major and minor axes of the fitted ellipse for dimension measurements gave more more accurate results than using the bounding rectangle dimensions. Through several analyses with differently shaped particles with different orientations, it was clearly determined that the dimension of fitted ellipse was a better option than the bounding rectangle for dimensions. However, shape-based correction factors were still needed for measurement of linear dimensions.
Secondly, shape parameters of particles of geometric shapes were also observed in order to develop the shape identification strategy employed by the plug-in. These shape parameters included reciprocal aspect ratio (RAR), rectangularity (RTY), and feret major axis ratio (FMR). Each gives specific values for different shapes depending on their parameters, which were implemented into the plug-in to classify the shapes of particles.
The shape identification and dimension measurement accuracy of the plug-in was then tested using an image containing a group of known geometric shapes and dimensions. This test showed a high accuracy for the calculations of the areas but a significantly lower accuracy for the perimeter calculations. However, it was discussed that the significantly lower accuracy for the perimeter calculations was not of high importance since they can be determined other ways and are not the primary focus of the plug-in. It was also observed that as shapes got smaller and smaller, after a certain point they could only be represented by a straight square in the image. This would lead to some misclassifications with samples, as shapes were represented by less and less pixels. The conclusion was that further research would have to be made in order to avoid misclassifications associated with (1) overlapping shape parameters values, (2) smaller shapes being represented by few pixels, (3) irregular shapes that deviate from the test cases, and (4) overlapping particles.
After, the effect of particle shape, size, and orientation on the determined length and width were tested. The test for the effect of particle shape on the determined length and width showed small mean deviations across all shapes. The test for the effect of particle area (size) on the determined length and width showed that the deviations tended to decrease with an increase in particle area. Though the lengths of triangles and widths of ellipses showed increased deviation, this was not significant enough to show drastic effects on the determined length and width. Finally, the test for the effect of particle orientation on the determined length and width showed only minor variations with a gradual increasing trend with orientation angle.
Lastly, the plug-in was used to create a size distribution analysis with food grain and with ground biomass. Both tests proved the plug-in to be effective, and gave hope for its growing potential. ImageJ plug-ins have the power to provide tailor-made solutions to suit the specific needs of machine vision applications.