Modified DBSCAN Algorithm for Microscopic Image Analysis of Wood

Abstract

The analysis of the intern anatomy of wood samples for species identification is a complex task that only experts can perform accurately. Since there are not many experts in the world and their training can last decades, there is great interest in developing automatic processes to extract high-level information from microscopic wood images. The purpose of this work was to develop algorithms that could provide meaningful information for the classification process. The work focuses on hardwoods, which have a very diverse anatomy including many different features. The ray width is one of such features, with high diagnostic value, which is visible on the tangential section. A modified distance function for the DBSCAN algorithm was developed to identify clusters that represent rays, in order to count the number of cells in width. To test both the segmentation and the modified DBSCAN algorithms, 20 images were manually segmented, obtaining an average Jaccard index of 0.66 for the segmentation and an average index M = 0.78 for the clustering task. The final ray count had an accuracy of 0.91.

Computer Vision-Based Wood Identification: A Review

“…Machine learning operates primarily as software that recognises patterns from input images that are processed to define a descriptive structure to which the unknown image will be referenced [68]. This involves various stages, as follows.…”