Comparison of Local Feature Detectors and Descriptors for Visual Object Categorization

Local features are used in many computer vision tasks including visual object categorization, content-based image retrieval and object recognition to mention a few. Local features are points, blobs or regions in images that are extracted using a local feature detector. To make use of extracted local features the localized interest points are described using a local feature descriptor. A descriptor histogram vector is a compact representation of an image and can be used for searching and matching images in databases. In this thesis the performance of local feature detectors and descriptors is evaluated for object class detection task. Features are extracted from image samples belonging to several object classes. Matching features are then searched using random image pairs of a same class. The goal of this thesis is to find out what are the best detector and descriptor methods for such task in terms of detector repeatability and descriptor matching rate.

Modeling of fractal antennas

In this Master Thesis the characteristics of the chosen fractal microstrip antennas are investigated. For modeling has been used the structure of the square Serpinsky fractal curves. During the elaboration of this Master thesis the following steps were undertaken: 1) calculation and simulation of square microstrip antennа, 2) optimizing for obtaining the required characteristics on the frequency 2.5 GHz, 3) simulation and calculation of the second and third iteration of the Serpinsky fractal curves, 4) radiation patterns and intensity distribution of these antennas. In this Master’s Thesis the search for the optimal position of the port and fractal elements was conducted. These structures can be used in perspective for creation of antennas working at the same time in different frequency range.

Coarse Global Visual Localization of a Mobile Robot

Localization, which is the ability of a mobile robot to estimate its position within its environment, is a key capability for autonomous operation of any mobile robot. This thesis presents a system for indoor coarse and global localization of a mobile robot based on visual information. The system is based on image matching and uses SIFT features as natural landmarks. Features extracted from training images arestored in a database for use in localization later. During localization an image of the scene is captured using the on-board camera of the robot, features are extracted from the image and the best match is searched from the database. Feature matching is done using the k-d tree algorithm. Experimental results showed that localization accuracy increases with the number of training features used in the training database, while, on the other hand, increasing number of features tended to have a negative impact on the computational time. For some parts of the environment the error rate was relatively high due to a strong correlation of features taken from those places across the environment.

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.

Paper Surface Image Processing for Roughness Quality Measurement

Diplomityössä on käsitelty paperin pinnankarkeuden mittausta, joka on keskeisimpiä ongelmia paperimateriaalien tutkimuksessa. Paperiteollisuudessa käytettävät mittausmenetelmät sisältävät monia haittapuolia kuten esimerkiksi epätarkkuus ja yhteensopimattomuus sileiden papereiden mittauksissa, sekä suuret vaatimukset laboratorio-olosuhteille ja menetelmien hitaus. Työssä on tutkittu optiseen sirontaan perustuvia menetelmiä pinnankarkeuden määrittämisessä. Konenäköä ja kuvan-käsittelytekniikoita tutkittiin karkeilla paperipinnoilla. Tutkimuksessa käytetyt algoritmit on tehty Matlab® ohjelmalle. Saadut tulokset osoittavat mahdollisuuden pinnankarkeuden mittaamiseen kuvauksen avulla. Parhaimman tuloksen perinteisen ja kuvausmenetelmän välillä antoi fraktaaliulottuvuuteen perustuva menetelmä.

Reconstruction and analysis of surface variation using photometric stereo

In many industrial applications, accurate and fast surface reconstruction is essential for quality control. Variation in surface finishing parameters, such as surface roughness, can reflect defects in a manufacturing process, non-optimal product operational efficiency, and reduced life expectancy of the product. This thesis considers reconstruction and analysis of high-frequency variation, that is roughness, on planar surfaces. Standard roughness measures in industry are calculated from surface topography. A fast and non-contact method to obtain surface topography is to apply photometric stereo in the estimation of surface gradients and to reconstruct the surface by integrating the gradient fields. Alternatively, visual methods, such as statistical measures, fractal dimension and distance transforms, can be used to characterize surface roughness directly from gray-scale images. In this thesis, the accuracy of distance transforms, statistical measures, and fractal dimension are evaluated in the estimation of surface roughness from gray-scale images and topographies. The results are contrasted to standard industry roughness measures. In distance transforms, the key idea is that distance values calculated along a highly varying surface are greater than distances calculated along a smoother surface. Statistical measures and fractal dimension are common surface roughness measures. In the experiments, skewness and variance of brightness distribution, fractal dimension, and distance transforms exhibited strong linear correlations to standard industry roughness measures. One of the key strengths of photometric stereo method is the acquisition of higher frequency variation of surfaces. In this thesis, the reconstruction of planar high-frequency varying surfaces is studied in the presence of imaging noise and blur. Two Wiener filterbased methods are proposed of which one is optimal in the sense of surface power spectral density given the spectral properties of the imaging noise and blur. Experiments show that the proposed methods preserve the inherent high-frequency variation in the reconstructed surfaces, whereas traditional reconstruction methods typically handle incorrect measurements by smoothing, which dampens the high-frequency variation.

Tuotantofilosofiat

Työssä esitellään käytetyimpiä tuotantofilosofioita. Tuotantofilosofia on hyvin laaja käsite ja sen vuoksi myös jotkin esiteltävistä menetelmistä ovat hyvin kaukana toisistaan. Työ koostuu teoriaosiosta, jossa on esitelty kukin tuotantofilosofia ja lopuksi johtopäätöksiä-osiossa käsitellään sitä, kuinka menetelmät liittyvät toisiinsa. Työssä esitellään JIT/JOT-tuotanto, Lean-tuotanto, Monozukuri, Modulointi, Standardointi, Strategiatyö, Six Sigma, TQM, TPM, QFD, MFD, Simulointi, Digitaalinen valmistus, DFX ja ns. uudet tuotantofilosofiat. Eri menetelmistä löytyvää lähdemateriaalia on tarjolla monipuolisesti, josta johtuen menetelmistä on voitu esitellä vain pääpiirteet. Tuotantofilosofioiden avulla voidaan saavuttaa monia eri asioita. Osa menetelmistä on luotu tuotannon tehostamiseksi ja yksinkertaistamiseksi, osa puolestaan lisää tuotannon tai koko yrityksen laatutasoa ja osa puolestaan helpottaa tuotteiden suunnittelu-työtä. Moni esiteltävistä filosofioista ei istu yksinomaan yhteen edellä mainituista kategorioista vaan kattaa laajempia alueita pitäen sisällään jopa kaikkia kolmea mainittua tulosta. Näiden lisäksi työssä on esitelty lyhyesti uusia tuotantofilosofioita, jotka ovat hieman irrallisia kokonaisuuksia verrattuna muihin työssä esiteltäviin filosofioihin. Työn tarkoituksena on auttaa hahmottamaan suurta kokonaisuutta jonka tuotantofilosofiat tuottavat. On tärkeää osata hahmottaa filosofioiden riippuvuus toisistaan ja se, että otettaessa käyttöön jotain tuotantofilosofiaa, tarkoittaa se myös mahdollisesti monen muunkin asian huomioonottamista. Tätä näkökantaa selvennetään johtopäätöksissä.

Pencil grip : a descriptive model and four empirical studies

Previous studies on pencil grip have typically dealt with the developmental aspects in young children while handwriting research is mainly concerned with speed and legibility. Studies linking these areas are few. Evaluation of the existing pencil grip studies is hampered by methodological inconsistencies. The operational definitions of pencil grip arerational but tend to be oversimplified while detailed descriptors tend to be impractical due to their multiplicity. The present study introduces a descriptive two-dimensional model for the categorisation of pencil grip suitable for research applications in a classroom setting. The model is used in four empirical studies of children during the first six years of writing instruction. Study 1 describes the pencil grips observed in a large group of pupils in Finland (n = 504). The results indicate that in Finland the majority of grips resemble the traditional dynamic tripod grip. Significant genderrelated differences in pencil grip were observed. Study 2 is a longitudinal exploration of grip stability vs. change (n = 117). Both expected and unexpected changes were observed in about 25 per cent of the children's grips over four years. A new finding emerged using the present model for categorisation: whereas pencil grips would change, either in terms of ease of grip manipulation or grip configuration, no instances were found where a grip would have changed concurrently on both dimensions. Study 3 is a cross-cultural comparison of grips observed in Finland and the USA (n = 793). The distribution of the pencil grips observed in the American pupils was significantly different from those found in Finland. The cross-cultural disparity is most likely related to the differences in the onset of writing instruction. The differences between the boys' and girls' grips in the American group were non-significant.An implication of Studies 2 and 3 is that the initial pencil grip is of foremost importance since pencil grips are largely stable over time. Study 4 connects the pencil grips to assessment of the mechanics of writing (n = 61). It seems that certain previously not recommended pencil grips might nevertheless be includedamong those accepted since they did not appear to hamper either fluency or legibility.

Local Binary Patterns in Focal-Plane Processing. Analysis and Applications

Feature extraction is the part of pattern recognition, where the sensor data is transformed into a more suitable form for the machine to interpret. The purpose of this step is also to reduce the amount of information passed to the next stages of the system, and to preserve the essential information in the view of discriminating the data into different classes. For instance, in the case of image analysis the actual image intensities are vulnerable to various environmental effects, such as lighting changes and the feature extraction can be used as means for detecting features, which are invariant to certain types of illumination changes. Finally, classification tries to make decisions based on the previously transformed data. The main focus of this thesis is on developing new methods for the embedded feature extraction based on local non-parametric image descriptors. Also, feature analysis is carried out for the selected image features. Low-level Local Binary Pattern (LBP) based features are in a main role in the analysis. In the embedded domain, the pattern recognition system must usually meet strict performance constraints, such as high speed, compact size and low power consumption. The characteristics of the final system can be seen as a trade-off between these metrics, which is largely affected by the decisions made during the implementation phase. The implementation alternatives of the LBP based feature extraction are explored in the embedded domain in the context of focal-plane vision processors. In particular, the thesis demonstrates the LBP extraction with MIPA4k massively parallel focal-plane processor IC. Also higher level processing is incorporated to this framework, by means of a framework for implementing a single chip face recognition system. Furthermore, a new method for determining optical flow based on LBPs, designed in particular to the embedded domain is presented. Inspired by some of the principles observed through the feature analysis of the Local Binary Patterns, an extension to the well known non-parametric rank transform is proposed, and its performance is evaluated in face recognition experiments with a standard dataset. Finally, an a priori model where the LBPs are seen as combinations of n-tuples is also presented