A new versatile algorithm to extract particle’s features from FESEM images: method evaluation and a case study on rice kernels

Image analysis approaches allow to quickly extract important information from images of diverse nature. Many techniques produce as a result images that contain regular and irregular objects. The ability of automatically extracting the objects and their related morphological features and properties is becoming fundamental, especially when the number of images to analyse is consistent. In this context, a new algorithm able to extract a series of morphological features from FESEM images was developed. Starting from a case study on 54 varieties of rice kernels, 220 images were acquired, and the algorithm was coded with the aim of extracting information from the round-shaped starch particles naturally present in rice kernels. The algorithm constitutes of different steps to segment the images and identify the object shapes and boundaries. Once those objects are identified, the algorithm extracts their morphological features, the number of identified objects and the amount of empty spaces among those objects. The developed algorithm is suitable for a rapid and automated analysis of several images, with the aim of extracting object-related morphological features and information about the general objects space disposition. The use of adaptive thresholds and correction steps allow to analyse images of different natures containing also defective and non-representative objects that will be automatically removed from the features calculation. In addition, to evaluate the algorithm performances, a Design of Experiment approach was developed to determine the effect of the input parameters choice on the algorithm output results, highlighting which parameters show a stronger effect on the output. • FESEM image analysis to automatically extract morphological features. • Tested on 220 images of 54 rice varieties, to analyse starch particles. • Use of adaptive thresholds and parameters to improve precision in diverse images. • DoE approach to assess the impact of input parameters on results and performance.