Multi-scale Image Segmentation Algorithm Research Articles

Enhancing the Thermal Images of the Upper Scarp of the Poggio Baldi Landslide (Italy) by Physical Modeling and Image Analysis

We present new methods for physical interpretation and mathematical treatment of the imaging contrast observed in thermal images of the rocky upper scarp of the Poggio Baldi landslide (Italy), which is part of a natural laboratory. Exemplar thermal images have been acquired with a high-performance camera at a distance of around 500 m, in a geometry where reflection is expected to dominate over thermal emission. The digital pixel intensities have therefore been considered as wavelength-integrated infrared spectral reflectance, irrespective of the temperature scale loaded into the camera software. Sub-portions of the scarp producing a lower signal have been identified by a multiscale image segmentation algorithm and overlaid on the visible image to provide an interpretation for the different thermal imaging contrast mechanisms that may be exploited for landslide monitoring in the future.

The Density Fraction Estimation of Coarse Coal by Use of the Kernel Method and Machine Vision

Coal density distribution is one of the most important production indexes in coal preparation processes, but its traditional measure is time consuming. A fast and efficient method to estimate the density fraction of coal particles becomes an urgent problem. This article proposed a prediction method of density fractions of coarse coal by use of kernel methods and machine vision. Coal particle images were segmented and identified by a multi-scale image segmentation algorithm based on the Hessian Matrix. Thirty-two features, including gray information and texture information, were extracted. The 3σ principle in statistics was applied to remove the abnormal points existing in features of each density fraction, and then all features were normalized. Kernel principal component analysis was used to reduce the data dimensions and the first two principle components were determined as the input of the support vector machine to predict the density fractions of coal particles. The best support vector machine parameters of c and g were determined by the method of K-fold cross validation. Through five tests, the prediction accuracy of training data reaches 84.67%, and the prediction accuracy of test data reaches 81.2%. Results indicated that the kernel method of kernel principal component analysis and support vector machine is able to predict the density fractions of overlap coarse coal.

Application of MultiScale Hidden Markov Modeling Wavelet Coefficients to fMRI Activation Detection

Problem Statement: The problem of detection of functional magnetic resonance images (fMRIs), that is, to decide active and nonactive regions of human brain from fMRIs is studied in this paper. fMRI research is finding and will find more and more applications in diagnosing and treating brain diseases like depression and schizophrenia. At its initial stage fMRI detection are pixel-wise methods, which do not take advantage of mutual information among neighboring pixels. Ignoring such spatial information can reduce detection accuracy. During past decade, many efforts have been focusing on taking advantage of spatial correlation inherent in fMRI data. Most well known is smoothing using a fixed Gaussian filter and the compensation for multiple testing using Gaussian random field theory as used by Statistical Parametric Mapping (SPM). Other methods including wavelets had also been proposed by the community. Approach: In this study a novel two-step approach was put forward that incorporates spatial correlation information and is amenable to analysis and optimization. First, a new multi scale image segmentation algorithm was proposed to decompose the correlation image into several different regions, each of which is of homogeneous statistical behavior. Second, each region will be classified independently as active or inactive using existing pixel-wise test methods. The image segmentation consists of two procedures: Edge detection followed by label estimation. To deduce the presence or absence of an edge from continuous data, two fundamental assumption of our algorithm are 1) each wavelet coefficient was described by a 2-state Gaussian Mixture Model (GMM); 2) across scale, each state is caused by its parent state, hence the Multiscale Hidden Markov Model (MHMM). The states of Markov chain are unknown (hidden) and represent the presence (state 1) or absence (state 0) of edges. Using this interpretation, the edge detection problem boils down to the posterior state estimation given observation. Results: Data processing results demonstrate much improved efficiency of our algorithm compared with pixel-wise detection methods. Conclusions: Our methods and results stress the importance of spatial-temporal modeling in fMRI research.