Generative Part-Based Gabor Object Detector

Object detection is a fundamental task of computer vision that is utilized as a core part in a number of industrial and scientific applications, for example, in robotics, where objects need to be correctly detected and localized prior to being grasped and manipulated. Existing object detectors vary in (i) the amount of supervision they need for training, (ii) the type of a learning method adopted (generative or discriminative) and (iii) the amount of spatial information used in the object model (model-free, using no spatial information in the object model, or model-based, with the explicit spatial model of an object). Although some existing methods report good performance in the detection of certain objects, the results tend to be application specific and no universal method has been found that clearly outperforms all others in all areas. This work proposes a novel generative part-based object detector. The generative learning procedure of the developed method allows learning from positive examples only. The detector is based on finding semantically meaningful parts of the object (i.e. a part detector) that can provide additional information to object location, for example, pose. The object class model, i.e. the appearance of the object parts and their spatial variance, constellation, is explicitly modelled in a fully probabilistic manner. The appearance is based on bio-inspired complex-valued Gabor features that are transformed to part probabilities by an unsupervised Gaussian Mixture Model (GMM). The proposed novel randomized GMM enables learning from only a few training examples. The probabilistic spatial model of the part configurations is constructed with a mixture of 2D Gaussians. The appearance of the parts of the object is learned in an object canonical space that removes geometric variations from the part appearance model. Robustness to pose variations is achieved by object pose quantization, which is more efficient than previously used scale and orientation shifts in the Gabor feature space. Performance of the resulting generative object detector is characterized by high recall with low precision, i.e. the generative detector produces large number of false positive detections. Thus a discriminative classifier is used to prune false positive candidate detections produced by the generative detector improving its precision while keeping high recall. Using only a small number of positive examples, the developed object detector performs comparably to state-of-the-art discriminative methods.

Supervised localimage feature detection

This thesis is about detection of local image features. The research topic belongs to the wider area of object detection, which is a machine vision and pattern recognition problem where an object must be detected (located) in an image. State-of-the-art object detection methods often divide the problem into separate interest point detection and local image description steps, but in this thesis a different technique is used, leading to higher quality image features which enable more precise localization. Instead of using interest point detection the landmark positions are marked manually. Therefore, the quality of the image features is not limited by the interest point detection phase and the learning of image features is simplified. The approach combines both interest point detection and local description into one phase for detection. Computational efficiency of the descriptor is therefore important, leaving out many of the commonly used descriptors as unsuitably heavy. Multiresolution Gabor features has been the main descriptor in this thesis and improving their efficiency is a significant part. Actual image features are formed from descriptors by using a classifierwhich can then recognize similar looking patches in new images. The main classifier is based on Gaussian mixture models. Classifiers are used in one-class classifier configuration where there are only positive training samples without explicit background class. The local image feature detection method has been tested with two freely available face detection databases and a proprietary license plate database. The localization performance was very good in these experiments. Other applications applying the same under-lying techniques are also presented, including object categorization and fault detection.

Rigid and Non-rigid Point-based Medical Image Registration

The primary goal of this dissertation is to develop point-based rigid and non-rigid image registration methods that have better accuracy than existing methods. We first present point-based PoIRe, which provides the framework for point-based global rigid registrations. It allows a choice of different search strategies including (a) branch-and-bound, (b) probabilistic hill-climbing, and (c) a novel hybrid method that takes advantage of the best characteristics of the other two methods. We use a robust similarity measure that is insensitive to noise, which is often introduced during feature extraction. We show the robustness of PoIRe using it to register images obtained with an electronic portal imaging device (EPID), which have large amounts of scatter and low contrast. To evaluate PoIRe we used (a) simulated images and (b) images with fiducial markers; PoIRe was extensively tested with 2D EPID images and images generated by 3D Computer Tomography (CT) and Magnetic Resonance (MR) images. PoIRe was also evaluated using benchmark data sets from the blind retrospective evaluation project (RIRE). We show that PoIRe is better than existing methods such as Iterative Closest Point (ICP) and methods based on mutual information. We also present a novel point-based local non-rigid shape registration algorithm. We extend the robust similarity measure used in PoIRe to non-rigid registrations adapting it to a free form deformation (FFD) model and making it robust to local minima, which is a drawback common to existing non-rigid point-based methods. For non-rigid registrations we show that it performs better than existing methods and that is less sensitive to starting conditions. We test our non-rigid registration method using available benchmark data sets for shape registration. Finally, we also explore the extraction of features invariant to changes in perspective and illumination, and explore how they can help improve the accuracy of multi-modal registration. For multimodal registration of EPID-DRR images we present a method based on a local descriptor defined by a vector of complex responses to a circular Gabor filter.

Unsupervised segmentation using Gabor wavelets and statistical features in LIDAR data analysis

In this paper, we address issues in segmentation Of remotely sensed LIDAR (LIght Detection And Ranging) data. The LIDAR data, which were captured by airborne laser scanner, contain 2.5 dimensional (2.5D) terrain surface height information, e.g. houses, vegetation, flat field, river, basin, etc. Our aim in this paper is to segment ground (flat field)from non-ground (houses and high vegetation) in hilly urban areas. By projecting the 2.5D data onto a surface, we obtain a texture map as a grey-level image. Based on the image, Gabor wavelet filters are applied to generate Gabor wavelet features. These features are then grouped into various windows. Among these windows, a combination of their first and second order of statistics is used as a measure to determine the surface properties. The test results have shown that ground areas can successfully be segmented from LIDAR data. Most buildings and high vegetation can be detected. In addition, Gabor wavelet transform can partially remove hill or slope effects in the original data by tuning Gabor parameters.

Face verification using Gabor wavelets and AdaBoost

This paper presents a new face verification algorithm based on Gabor wavelets and AdaBoost. In the algorithm, faces are represented by Gabor wavelet features generated by Gabor wavelet transform. Gabor wavelets with 5 scales and 8 orientations are chosen to form a family of Gabor wavelets. By convolving face images with these 40 Gabor wavelets, the original images are transformed into magnitude response images of Gabor wavelet features. The AdaBoost algorithm selects a small set of significant features from the pool of the Gabor wavelet features. Each feature is the basis for a weak classifier which is trained with face images taken from the XM2VTS database. The feature with the lowest classification error is selected in each iteration of the AdaBoost operation. We also address issues regarding computational costs in feature selection with AdaBoost. A support vector machine (SVM) is trained with examples of 20 features, and the results have shown a low false positive rate and a low classification error rate in face verification.

Facial feature extraction and selection by Gabor wavelets and boosting

In this paper, we present a feature selection approach based on Gabor wavelet feature and boosting for face verification. By convolution with a group of Gabor wavelets, the original images are transformed into vectors of Gabor wavelet features. Then for individual person, a small set of significant features are selected by the boosting algorithm from a large set of Gabor wavelet features. The experiment results have shown that the approach successfully selects meaningful and explainable features for face verification. The experiments also suggest that for the common characteristics such as eyes, noses, mouths may not be as important as some unique characteristic when training set is small. When training set is large, the unique characteristics and the common characteristics are both important.

On the use of textural features for writer identification in old handwritten music scores

Writer identification consists in determining the writer of a piece of handwriting from a set of writers. In this paper we present a system for writer identification in old handwritten music scores which uses only music notation to determine the author. The steps of the proposed system are the following. First of all, the music sheet is preprocessed for obtaining a music score without the staff lines. Afterwards, four different methods for generating texture images from music symbols are applied. Every approach uses a different spatial variation when combining the music symbols to generate the textures. Finally, Gabor filters and Grey-scale Co-ocurrence matrices are used to obtain the features. The classification is performed using a k-NN classifier based on Euclidean distance. The proposed method has been tested on a database of old music scores from the 17th to 19th centuries, achieving encouraging identification rates.

Spatial coherence does not affect contrast discrimination for multiple Gabor stimuli

Gestalt grouping rules imply a process or mechanism for grouping together local features of an object into a perceptual whole. Several psychophysical experiments have been interpreted as evidence for constrained interactions between nearby spatial filter elements and this has led to the hypothesis that element linking might be mediated by these interactions. A common tacit assumption is that these interactions result in response modulation which disturbs a local contrast code. We addressed this possibility by performing contrast discrimination experiments using two-dimensional arrays of multiple Gabor patches arranged either (i) vertically, (ii) in circles (coherent conditions), or (iii) randomly (incoherent condition), as well as for a single Gabor patch. In each condition, contrast increments were applied to either the entire test stimulus (experiment 1) or a single patch whose position was cued (experiment 2). In experiment 3, the texture stimuli were reduced to a single contour by displaying only the central vertical strip. Performance was better for the multiple-patch conditions than for the single-patch condition, but whether the multiple-patch stimulus was coherent or not had no systematic effect on the results in any of the experiments. We conclude that constrained local interactions do not interfere with a local contrast code for our suprathreshold stimuli, suggesting that, in general, this is not the way in which element linking is achieved. The possibility that interactions are involved in enhancing the detectability of contour elements at threshold remains unchallenged by our experiments.

Features combination for art authentication studies: brushstroke and materials analysis of Amadeo de Souza-Cardoso

This work presents a tool to support authentication studies of paintings attributed to the modernist Portuguese artist Amadeo de Souza-Cardoso (1887-1918). The strategy adopted was to quantify and combine the information extracted from the analysis of the brushstroke with information on the pigments present in the paintings. The brushstroke analysis was performed combining Gabor filter and Scale Invariant Feature Transform. Hyperspectral imaging and elemental analysis were used to compare the materials in the painting with those present in a database of oil paint tubes used by the artist. The outputs of the tool are a quantitative indicator for authenticity, and a mapping image that indicates the areas where materials not coherent with Amadeo's palette were detected, if any. This output is a simple and effective way of assessing the results of the system. The method was tested in twelve paintings obtaining promising results.

Tubular Variant Of Basal Cell Adenoma Shares Immunophenotypical Features With Normal Intercalated Ducts And Is Closely Related To Intercalated Duct Lesions Of Salivary Gland.

The morphological criteria for identification of intercalated duct lesions (IDLs) of salivary glands have been defined recently. It has been hypothesised that IDL could be a precursor of basal cell adenoma (BCA). BCAs show a variety of histological patterns, and the tubular variant is the one that presents the strongest resemblance with IDLs. The aim of this study was to analyse the morphological and immunohistochemical profiles of IDLs and BCAs classified into tubular and non-tubular subtypes, to determine whether or not IDL and tubular BCA represent distinct entities. Eight IDLs, nine tubular BCAs and 19 non-tubular BCAs were studied. All tubular BCAs contained IDL-like areas, which represented 20-70% of the tumour. In non-tubular BCA, IDL-like areas were occasional and small (<5%). One patient presented IDLs, tubular BCAs and IDL/tubular BCA combined lesions. Luminal ductal cells of IDLs and tubular BCAs exhibited positivity for CK7, lysozyme, S100 and DOG1. In the non-tubular BCA group, few luminal cells exhibited such an immunoprofile; they were mainly CK14-positive. Basal/myoepithelial cells of IDLs, tubular BCAs and non-tubular BCAs were positive for CK14, calponin, α-SMA and p63; they were more numerous in BCA lesions. IDL, tubular BCA and non-tubular BCA form a continuum of lesions in which IDLs are related closely to tubular BCA. In both, the immunoprofile of luminal and myoepithelial cells recapitulates the normal intercalated duct. The difference between the adenoma-like subset of IDLs and tubular BCA rests mainly on the larger numbers of myoepithelial cells in the latter. Our findings indicate that at least some BCAs can arise via IDLs.

[demographic And Clinical Features Of Patients With Rheumatoid Arthritis In Piauí, Brazil--evaluation Of 98 Patients].

Brazilian epidemiological studies on rheumatoid arthritis are scarce, mainly in the northeast; thus many data currently available originate from the international literature. To describe demographic, clinical and serological characteristics of patients with rheumatoid arthritis (RA) followed-up by the same physician, in state of Piauí, Brazil. Data were collected between August 2010 and March 2013, in three health services of Piauí that provided health care in Rheumatology: a university-affiliated hospital, a public outpatient clinic and a private clinic. The numbers represent mean ± SD or percentage: 47.5±11.03 years-old non-Caucasian woman, non-smoker (59.2%), low educational level, mean disease duration of 7.7 years ± 7.6, and major extra-articular manifestations were rheumatoid nodules (19.4%) and sicca syndrome (46.9%). Features of rheumatoid arthritis obtained in this study are similar to those found in some national and international studies, but we observed higher female preponderance and illiteracy rate, in addition to a moderately severe erosive disease on average, with frequent sicca and other extra-articular manifestations.

Laser Printer Attribution: Exploring New Features And Beyond.

With a huge amount of printed documents nowadays, identifying their source is useful for criminal investigations and also to authenticate digital copies of a document. In this paper, we propose novel techniques for laser printer attribution. Our solutions do not need very high resolution scanning of the investigated document and explore the multidirectional, multiscale and low-level gradient texture patterns yielded by printing devices. The main contributions of this work are: (1) the description of printed areas using multidirectional and multiscale co-occurring texture patterns; (2) description of texture on low-level gradient areas by a convolution texture gradient filter that emphasizes textures in specific transition areas and (3) the analysis of printer patterns in segments of interest, which we call frames, instead of whole documents or only printed letters. We show by experiments in a well documented dataset that the proposed methods outperform techniques described in the literature and present near-perfect classification accuracy being very promising for deployment in real-world forensic investigations.

Surgical Treatment Of Microinvasive Cervical Cancer: Analysis Of Pathologic Features With Implications On Radicality.

To evaluate pathologic features with implications on surgical radicality in women treated with radical hysterectomy and pelvic lymphadenectomy for cervical cancer stage IA1 with lymph vascular space invasion (LVSI) and stage IA2 by correlating findings in conization and hysterectomy specimens. Women with cervical cancer stage IA1 with LVSI and stage IA2 diagnosed by loop electrosurgical excisional procedure or cold knife conization were treated with radical hysterectomy and pelvic lymphadenectomy from January 1999 to December 2011 in 2 institutions. Fifty patients were enrolled: 40 with stage IA2 and 10 with stage IA1 with LVSI. Median age was 43 (30-67) years. All patients underwent cervical conization for diagnosis (45 loop electrosurgical excisional procedure, 5 cold knife). Lymph vascular space invasion was detected in 15 patients (30%). Two patients had positive pelvic nodes. No parametrial involvement was detected in the entire cohort. Positive margins were present in 35 patients, and residual disease was detected in 22 patients (44%). Positive margins predicted residual disease at radical hysterectomy (P = 0.02). Medium follow-up time was 51 months. One patient developed a pelvic recurrence, and there were no disease-related deaths. Patients with positive margins in cone biopsy specimens have an increased risk of residual disease at radical hysterectomy and require careful evaluation before conservative surgery. Pelvic lymph node evaluation is essential because lymph node metastasis may occur even in early stages. The lack of parametrial invasion in this study reinforces the knowledge that the select group of patients with microinvasive cervical carcinoma stages IA1 LVSI and stage IA2 have a very low risk of parametrial infiltration. Less radical surgery can be carefully considered for these patients.

Apert syndrome: clinical and radiographic features and case report

PURPOSE: Apert syndrome is a rare type I acrocephalosyndactyly syndrome characterized by craniosynostosis, severe syndactyly of the hands and feet, and dysmorphic facial features. Presents autosomal dominant inheritance assigned to mutations in the fibroblast growth factor receptors gene. The oral cavity of Apert patients includes a reduction in the size of the maxilla, tooth crowding, anterior open-bite of the maxilla, impacted teeth, delayed eruption, ectopic eruption, supernumerary teeth, and thick gingiva. The mandible usually is within normal size and shape, and simulates a pseudoprognathism. CASE DESCRIPTION: A female patient, 13 years old, with diagnosis of Apert syndrome, attended a dental radiology clinic. The clinical signs were occular anomalies, dysmorphic facial features, syndactyly and oral features observed clinically and radiographically. The patient was referred to a specialized center of clinical care for patients with special needs. CONCLUSION: Because of the multiple alterations in patients with Apert syndrome, a multidisciplinary approach, including dentists and neurosurgeons, plastic surgeons, ophthalmologists and geneticists, is essential for a successful planning and treatment.

Morpho-anatomical features of underground systems in six Asteraceae species from the Brazilian Cerrado

In the Brazilian Cerrado (neotropical savanna), the development of bud-bearing underground systems as adaptive structures to fire and dry periods can comprise an important source of buds for this ecosystem, as already demonstrated in the Brazilian Campos grasslands and North American prairies. Asteraceae species from both woody and herbaceous strata have subterranean organs that accumulate carbohydrates, reinforcing the adaptive strategy of these plants to different environmental conditions. This study aims to analyse the morpho-anatomy of underground systems of six species of Asteraceae (Mikania cordifolia L.f. Willd., Mikania sessilifolia DC, Trixis nobilis (Vell.) Katinas, Pterocaulon alopecuroides (Lam.) DC., Vernonia elegans Gardner and Vernonia megapotamica Spreng.), to describe these structures and to verify the occurrence and origin of shoot buds, and to analyse the presence of reserve substances. Individuals sampled in Cerrado areas in São Paulo State showed thick underground bud-bearing organs, with adventitious or lateral roots and presence of fructans. Xylopodium was found in all studied species, except for Trixis nobilis, which had stem tuber. The presence of fructans as reserve, and the capacity of structures in the formation of buds indicate the potential of herbaceous species of Asteraceae in forming a viable bud bank for vegetation regeneration in the Brazilian Cerrado.