Image-Based Machine Learning for Reduction of User Fatigue in an Interactive Model Calibration System

The interactive multiobjective genetic algorithm (IMOGA) is a promising new approach to calibrate models. The IMOGA combines traditional optimization with an interactive framework, thus allowing both quantitative calibration criteria as well as the subjective knowledge of experts to drive the search for model parameters. One of the major challenges in using such interactive systems is the burden they impose on the experts that interact with the system. This paper proposes the use of a novel image-based machine-learning (IBML) approach to reduce the number of user interactions required to identify promising calibration solutions involving spatially distributed parameter fields (e.g., hydraulic conductivity parameters in a groundwater model). The first step in the IBML approach involves selecting a few highly representative solutions for expert ranking. The selection is performed using unsupervised clustering approaches from the field of image processing, which group potential parameter fields based on their spatial similarities. The expert then ranks these representative solutions, after which a machine-learning model (augmented with the spatial information of the selected fields) is trained to learn user preferences and predict rankings for solutions not ranked by the expert. To better mimic the “visual” information processing of human experts, algorithms from the field of image processing are used to mine information about the spatial characteristics of parameter fields, thus improving the performance of the clustering and machine-learning algorithms. The IBML approach is tested and demonstrated on a groundwater calibration problem and is shown to lead to significant improvements, reducing the amount of user interaction by as much as half without compromising the solution quality of the IMOGA.     

Gridline: automatic grid alignment DNA microarray scans

We present a new automatic grid alignment algorithm for detecting two-dimensional (2-D) arrays of spots in DNA microarray images. Our motivation for this work is the lack of automation in high-throughput microarray data analysis that leads to a) spatial inaccuracy of located spots and hence inaccuracy of extracted information from a spot and b) inconsistency of extracted features due to manual selection of grid alignment parameters. The proposed grid alignment algorithm is novel in the sense that 1) it can detect irregularly row- and column-spaced spots in a 2-D array, 2) it is independent of spot color and size, 3) it is general to localize a grid of other primitive shapes than the spot shapes, 4) it can perform grid alignment on any number of input channels, 5) it reduces the number of free parameters to minimum by data driven optimization of most algorithmic parameters, and 6) it has a built-in speed versus accuracy tradeoff mechanism to accommodate user's requirements on performance time and accuracy of the results. The developed algorithm also automatically identifies multiple blocks of 2-D arrays, as it is the case in microarray images, and compensates for grid rotations in addition to grid translations.     

Active vision for reliable ranging: Cooperating focus,stereo, and vergence

This article addresses the problem of measuring reliabily the absolute three-dimensional position of objects in an unknown and cluttered scene. It circumvents the limitations of a single sensor or single algorithm by using several range recovery techniques together, so that they cooperate in visual behaviors similar to those exhibited by the human visual system. Implemented visual behaviors include (i) aperture adjustment to vary depth of field and contrast, (ii) focus ranging followed by fixation, (iii) stereo ranging followed by focus ranging, and (iv) focus ranging followed by disparity prediction followed by focus ranging. The main contribution is a demonstration that two particular visual ranging processes—focusing and stereo—can cooperate to improve measurement reliability. The results of 75 experiments processing close to 3000 different object points lying at distances between 1 and 3 meters demonstrate that the computed range values are highly reliable.     

Report on workshop on high performance computing and communicationsfor grand challenge applications: computer vision, speech and naturallanguage processing, and artificial intelligence

The findings of a workshop, the goals of which were to identify applications, research problems, and designs of high performance computing and communications (HPCC) systems for supporting applications are discussed. In computer vision, the main scientific issues are machine learning, surface reconstruction, inverse optics and integration, model acquisition, and perception and action. In speech and natural language processing (SNLP), issues were identified statistical analysis in corpus-based speech and language understanding, search strategies for language analysis, auditory and vocal-tract modeling, integration of multiple levels of speech and language analyses, and connectionist systems. In AI, important issues that need immediate attention include the development of efficient machine learning and heuristic search methods that can adapt to different architectural configurations, and the design and construction of scalable and verifiable knowledge bases, active memories, and artificial neural networks     

Occlusions as a guide for planning the next view

A strategy for acquiring 3-D data of an unknown scene, using range images obtained by a light stripe range finder is addressed. The foci of attention are occluded regions, i.e., only the scene at the borders of the occlusions is modeled to compute the next move. Since the system has knowledge of the sensor geometry, it can resolve the appearance of occlusions by analyzing them. The problem of 3-D data acquisition is divided into two subproblems due to two types of occlusions. An occlusion arises either when the reflected laser light does not reach the camera or when the directed laser light does not reach the scene surface. After taking the range image of a scene, the regions of no data due to the first kind of occlusion are extracted. The missing data are acquired by rotating the sensor system in the scanning plane, which is defined by the first scan. After a complete image of the surface illuminated from the first scanning plane has been built, the regions of missing data due to the second kind of occlusions are located. Then, the directions of the next scanning planes for further 3-D data acquisition are computed     

Single-viewpoint, catadioptric cone mirror omnidirectional imaging theory and analysis

A family of catadioptric imaging systems has been developed that can achieve omnidirectional viewing with a single planar imager while still being able to recover perspective images, provided that they satisfy the single-viewpoint (SVP) constraint. It has been shown that the only mirror shapes that can have SVP when paired with a sole focusing planar imager camera are the surfaces of revolution of conic section curves. However, the special case of such a surface, the cone-shaped mirror itself, has not been deemed a viable SVP mirror shape. We present a comprehensive imaging theory of the cone mirror in its SVP configuration. We show that the SVP, cone mirror catadioptric system not only is practical but also has unique advantages for certain applications. The detailed theory explains why and how a practical SVP cone configuration can be set up, the merits and weaknesses of such systems, and how one can remedy the weaknesses to create a workable imaging system. We also derive the tolerance formula for estimating effects of alignment errors. A prototype has been constructed, and experimental results validate our theory.     

Location- and density-based hierarchical clustering usingsimilarity analysis

This paper presents a new approach to hierarchical clustering of point patterns. Two algorithms for hierarchical location- and density-based clustering are developed. Each method groups points such that maximum intracluster similarity and intercluster dissimilarity are achieved for point locations or point separations. Performance of the clustering methods is compared with four other methods. The approach is applied to a two-step texture analysis, where points represent centroid and average color of the regions in image segmentation     

A methodology for evaluation of task performance in robotic systems: a case study in vision-based localization

We investigated the performance of an agent that uses visual information in a partially unknown and changing environment in a principled way. We propose a methodology to study and evaluate the performance of autonomous agents. We first analyze the system theoretically to determine the most important system parameters and to predict error bounds and biases. We then conduct an empirical analysis to update and refine the model. The ultimate goal is to develop self-diagnostic procedures. We show that although simple models can successfully predict some major effects, empirically observed performance deviates from theoretical predictions in interesting ways.