Active learning is typically limited by the small sample problem which makes the resulting classifiers perform poorly,
especially in the initial stages. To overcome this problem, in this paper, a novel framework - graph-incorporated active
learning - is proposed, in which the selection pool is regarded as a graph. Its graph structure is applied to both improve
sample selection criterion and provide the learner enough pseudo-labeled samples. By comparing with the state-of-theart
technique, the experiments on benchmark datasets show that the improvement of the proposed method is significant,
i.e., it can solve the small problem well. The framework is combined with, but is not limited to, SVM.
With the rapid growth of image archives, many content-based image retrieval and annotation systems have been
developed for effectively indexing and searching these images. However, due to the semantic gap problem, these
systems are still far from satisfactory for practical use. Hence, bridging the semantic gap has been an area of intensive
research, in which several influential approaches that based upon an intermediate representation such as bag-of-words
(BOW) have demonstrated major successes. In most previous work,, the semantic context between visual words in BOW
is usually ignored or not exploited for the retrieval and annotation. To resolve this problem, we have developed a series
of approaches to semantic context extraction and representation that is based on the Markov models and kernel methods.
To our knowledge, this is the first application of kernel methods and 2D Markov models simultaneously to image
categorization and annotation which have been shown through experiments on standard benchmark datasets that they are
able to outperform several state-of-the-art methods.
In this paper, we present a scale independent automatic face location technique which can detect the locations of frontal
human faces from images. Our hierarchical approach of knowledge-based face detection composed of three levels. Level
1 consists of a simple but effective eyes model that generates a set of rules to judge whether or not there exists a human
face candidate in the current search area in a scale-independent manner and in a single scan of the image. To utilize this
model, we define a new operator - extended projection and define two new concepts: single projection line and pair
projection line. At level 2, an improved model of Yang's mosaic image model is applied to check the consistency of
visual features with respect to the human face within each 3x3 blocks of a candidate face image. At the third level, we
apply a SVM based face model, to eliminate the false positives obtained from level 2. Experimental results show the
combined rule-based and statistical approach works well in detecting frontal human faces in uncluttered scenes.
Many grey-level thresholding methods based on histogram or other statistic information about the interest image such
as maximum entropy and so on have been proposed in the past. However, most methods based on statistic analysis of the
images concerned little about the characteristics of morphology of interest objects, which sometimes could provide very
important indication which can help to find the optimum threshold, especially for those organisms which have special
texture morphologies such as vasculature, neuro-network etc. in medical imaging. In this paper, we propose a novel
method for thresholding the fluorescent vasculature image series recorded from Confocal Scanning Laser Microscope.
After extracting the basic orientation of the slice of vessels inside a sub-region partitioned from the images, we analysis
the intensity profiles perpendicular to the vessel orientation to get the reasonable initial threshold for each region. Then
the threshold values of those regions near the interest one both in x-y and optical directions have been referenced to get
the final result of thresholds of the region, which makes the whole stack of images look more continuous. The resulting
images are characterized by suppressing both noise and non-interest tissues conglutinated to vessels, while improving the
vessel connectivities and edge definitions. The value of the method for idealized thresholding the fluorescence images of
biological objects is demonstrated by a comparison of the results of 3D vascular reconstruction.
This paper addresses the problem of image content characterization in the compressed domain for the purpose of facilitating similarity matching in a multimedia database. Specifically, given the disparity of the content characterization power of compressed domain approaches and those based on pixel-domain features, with the latter being usually considered as the more superior one, our objective is to transform the selected set of compressed domain feature histograms in such a way that the retrieval result based on these features is compatible with their spatial domain counterparts. Since there are a large number of possible transformations, we adopt a genetic algorithm approach to search for the optimal one, where each of the binary strings in the population represents a candidate transformation. The fitness of each transformation is defined as a function of the discrepancies between the spatial-domain and compressed-domain retrieval results. In this way, the GA mechanism ensures that transformations which best approximate the performance of spatial domain retrieval will survive into the next generation and are allowed through the operations of crossover and mutation to generate variations of themselves to further improve their performances.
In this paper we present a semantic content representation scheme and the associated techniques for supporting (1) query by image examples or by natural language in a histological image database and (2) automatic annotation generation for images through image semantic analysis. In this research, various types of query are analyzed by either a semantic analyzer or a natural language analyzer to extract high level concepts and histological information, which are subsequently converted into an internal semantic content representation structure code-named 'Papillon.' Papillon serves not only as an intermediate representation scheme but also stores the semantic content of the image that will be used to match against the semantic index structure within the image database during query processing. During the image database population phase, all images that are going to be put into the database will go through the same processing so that every image would have its semantic content represented by a Papillon structure. Since the Papillon structure for an image contains high level semantic information of the image, it forms the basis of the technique that automatically generates textual annotation for the input images. Papillon bridges the gap between different media in the database, allows complicated intelligent browsing to be carried out efficiently, and also provides a well- defined semantic content representation scheme for different content processing engines developed for content-based retrieval.
This paper presents an approach for automatically assign histologically meaningful labels to tissue slide images. This approach is implemented as part of a larger system, I- Browse, which combines iconic and semantic content for intelligent image browsing. Our approach partitioned an input image into a number of subimages. A set of texture features based on Gabor filterings and color histogram which capture the visual characteristics of each of the subimages were computed. These image feature measurements then form the input to a pattern classifier which gives an initial coarse label assignment to subimages based on a hierarchical clustering of these image features. To facilitate supervised training of the classifier, a knowledge elicitation tool was developed which allows a histopathologist to assign histological terms to a sample of sub-images obtained from digitized tissue imags. The initial labels and their spatial distribution were then analyzed by a semantic analyzer with the help of a knowledge base which contains prior knowledge of the expected visual appearance of histological images of an organ. The label assigned to the subimages were successive refined through a process of relevant feedback.
Two sources of information play key roles in a collection of medical images such as computer tomographs, X-rays and histological slides, they are (1) textual descriptions relating to the image content and (2) visual features that can be seen on the image itself. The former are traditionally made by human specialists (e.g. histopathologists, radiographers, etc.) who interpret the image, and the latter are the inherent characteristics of images. This research program aims to study the architectural issues of a system which combines and interprets the information inherent in these two media to achieve automatic intelligent browsing of medical images. To give the research some practical significance, we applied the architecture to the design of the I-BROWSE system which is being developed jointly by the City University of Hong Kong and the Clinical School of the University of Cambridge. I- BROWSE is aimed to support intelligent retrieval and browsing of histological images obtained along the gastrointestinal tract (GI tract). Within such an architecture, given a query image or a populated image, a set of low level image feature measurements are obtained from a Visual Feature Detector, and with the help of knowledge bases and reasoning engines, the Semantic Analyzer derives, using an semantic feature generation and verification paradigm, the high level attributes for the image and furthermore automatically generates textual annotations for it. If the input image is accompanied with annotations made by a human specialist, the system will also analyze, combine and verify these two level of information, i.e., iconic and semantic contents. In the paper, we present the architectural issues and the strategies needed to support such information fusion process as well as the potentials of intelligent browsing using this dual- content-based approach.
This paper describes a method for automated assessment of the movement of the aorta from CT images using a new formulation of the active contour model to segment and track the aortic boundary and its movements through the temporal image sequence. The active contour model is capable of exploiting prior knowledge and posterior information of the image content, e.g. the anatomical shape of the aorta and the image gradient intensity of the aortic boundaries as well as physical constraint such as continuity of motion along the time dimension.
Many techniques in machine vision require tangent estimation. The direction-dependent tangent (DDT) is introduced. The representation makes explicit the direction of curve following and the concavity, hence facilitating shape matching. The scheme is a simple but powerful enhancement to the standard tangent notation. However, discrete tangent and curvature estimations are very sensitive to noise and quantization errors, and the original DDT formulation is difficult to apply directly to digital curves. Based on the geometric property of DDT, we propose to detect zero curvature points to determine the DDT values. Traditional approaches to zero curvature detection rely heavily on discrete tangent and curvature estimations, which are difficult to approximate accurately. Hence, most researchers adopted the multi-scale solutions, which are costly to compute. In this paper, we put forth a new measure, which we called `turning angle', for zero curvature detection. Based on this measure, we develop an efficient conditioning algorithm to tackle the zero curvature detection problem. Although the conditioning algorithm is quite straightforward, the zero curvature points detected are quite stable across scales.
Many techniques in machine vision require tangent estimation. In many implementations, the acquired tangent estimates are sensitive to coding direction of the curves in interest. Therefore, it is a common practice to enforce a certain coding scheme, for example, a boundary is traced in the counterclockwise manner. However, this scheme guarantees to work only for closed curves. For open curves, an inverse operator seems to be a must. In this paper, we propose a new tangent representation scheme named direction-dependent tangent (DDT). DDT makes explicit the direction of curve following and incorporates concavity information into the tangent orientation. The scheme is a simple but powerful enhancement to the standard tangent notation. It facilitates shape matching type of tasks by removing the need for either a predefined coding direction or an inverse operator.
This paper presents a real time inspection algorithm of fuel pellet surface. The algorithm identifies the different types of surface features of the pellets from sensed data derived from an array of proximity sensors. The algorithm meets the operating requirements in terms of speed and accuracy and can be implemented using relatively simple hardware. The development also includes a sensed data visualization tool which facilitates the analysis of the performance of the algorithm.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.