Automating shockwave segmentation in low-contrast coherent shadowgraphy

Abstract

The paper presents a method that enables automated segmentation of the low-contrast shadowgraph images, e.g., acquired in the studies of laser induced shockwave phenomena. The method is especially suitable for the analysis of large image data sets, such as obtained at studying the evolution of laser-induced shockwaves with high spatial and temporal resolution. The method comprises two active contours algorithms. First, the approximate shape of the shockwave is detected by a traditional snake algorithm using external energies that base on texture cues. The outcome of the coarse detection serves as an initialization to the second refining stage detection introducing a Greedy snake algorithm. Local optimum is searched with respect to responses of steerable filtering and edge orientation similarity by exploiting the Bayesian formalism. The paper presents validation of the method on sample of 12 low contrast shadowgraphs by comparison to the manual segmentation technique. The obtained results demonstrate overall good performance, robustness and high accuracy of the method.

Machine learning to analyze images of shocked materials for precise and accurate measurements

“…This method has not been widely used within the field, as it is not optimized for highnoise images. 26 Established image segmentation algorithms cannot give accurate results due to high noise content in the images and large overlap in pixel intensities between the different classes, i.e. the different shocked states.…”

Two-stage approach to extracting visual objects from paper documents

“…In most of them, the idea comes from the assumption that the first processing stage performs a rough detection of objects of interest, while the second one applies more precise means to improve the identification accuracy [25]. In many papers, the two-stage approach is related to the integration of features (e.g., appearance and spatio-temporal HOGs [26], difference-of-Gaussians and accumulated gradient projection vector [27], entropy of local histograms and heuristic features [28], edge information and SIFT features [29]), combining classifiers (e.g., SVM and random sample consensus-RANSAC [30], two stages of mean-shift clustering [31]), mixed approaches (e.g., Hough transform joined with DBSCAN clustering [32], edge map and SVM [33], HOG and SVM [34], two variants of snakes [35], particle swarm optimization and fuzzy classifier [36]). …”