Feature Extractionusing ORB-RANSAC for Face Recognition

Object recognition has a wide domain of applications such as content-based image classification, video data mining, video surveillance and more. Object recognition accuracy has been a significant concern. Although deep learning had automated the feature extraction but hand crafted features continue to deliver consistent performance. This paper aims at efficient object recognition using hand crafted features based on Oriented Fast & Rotated BRIEF (Binary Robust Independent Elementary Features) and Scale Invariant Feature Transform features. Scale Invariant Feature Transform (SIFT) are particularly useful for analysis of images in light of different orientation and scale. Locality Preserving Projection (LPP) dimensionality reduction algorithm is explored to reduce the dimensions of obtained image feature vector. The execution of the proposed work is tested by using k-NN, decision tree and random forest classifiers. A dataset of 8000 samples of 100-class objects has been considered for experimental work. A precision rate of 69.8% and 76.9% has been achieved using ORB and SIFT feature descriptors, respectively. A combination of ORB and SIFT feature descriptors is also considered for experimental work. The integrated technique achieved an improved precision rate of 85.6% for the same.

2020

In recent years there has been a great deal of research and study in the field of visual odometry, which has led to the development of practical processes such as visual based measurement in robotics and automotive technology. Direct methods, feature-based methods and hybrid methods are three common approaches in solving visual odometry problems and given the general belief that feature-based approach speeds are higher, this approach has been welcomed in recent years. Therefore, an attempt has been made in the present study to calculate the transformation matrix of two-dimensional sequential image sets using invariant features that can estimate the changes in camera rotation and translation. In the algorithm, two-steps of identifying keypoints and removing outliers are performed using five different local feature detection algorithms (SURF, SIFT, FAST, STAR, ORB) and RANdom SAmple Consensus algorithm (RANSAC), respectively. In addition, the impact of each of them, their intrinsic pa...

2011

We propose a fast local image feature detector and descriptor that is implementable on the GPU. Our method is the first GPU implementation of the popular FAST detector. A simple but novel method of feature orientation estimation which can be calculated in constant time is proposed. The robustness and reliability of our orientation estimation is validated against rotation invariant descriptors such as SIFT and SURF. Furthermore, we propose a binary feature descriptor which is robust to noise, scalable, rotation invariant, fast to compute in parallel and maintains low memory consumption. The proposed method demonstrates good robustness and very fast computation times, making it usable in real-time applications.

In this paper, we present a novel scale-and rotation-invariant interest point detector and descriptor, coined SURF (Speeded Up Robust Features). It approximates or even outperforms previously proposed schemes with respect to repeatability, distinctiveness, and robustness, yet can be computed and compared much faster. This is achieved by relying on integral images for image convolutions; by building on the strengths of the leading existing detectors and descriptors (in casu, using a Hessian matrix-based measure for the detector, and a distribution-based descriptor); and by simplifying these methods to the essential. This leads to a combination of novel detection, description, and matching steps. The paper presents experimental results on a standard evaluation set, as well as on imagery obtained in the context of a real-life object recognition application. Both show SURF's strong performance.

This article presents a novel scale-and rotation-invariant detector and descriptor, coined SURF (Speeded-Up Robust Features). SURF approximates or even outperforms previously proposed schemes with respect to repeatability, distinctiveness, and robustness, yet can be computed and compared much faster. This is achieved by relying on integral images for image convolutions; by building on the strengths of the leading existing detectors and descriptors (specifically, using a Hessian matrix-based measure for the detector, and a distribution-based descriptor); and by simplifying these methods to the essential. This leads to a combination of novel detection, description, and matching steps.

2017 Artificial Intelligence and Signal Processing Conference (AISP), 2017

This paper describes the method to increase the speed of SIFT feature extraction by feature approximation instead of feature calculation in various layers. SIFT has been proven to be the most robust local rotation and illumination invariant feature descriptor. Additionally, it supports affine transformation. Being fully scale invariant is the most important advantage of this descriptor. The most major SIFT'sdrawback is time-consuming which prevents utilizing SIFTin real time applications. This research attempts to decrease computational cost withoutsacrificing performance. The recent researches in this area approved that direct feature computation is more expensive than extrapolation. Consequently, contribution of this research reduces processing time considerablywithout losing accuracy.

Proceedings of the 27th Conference on Image and Vision Computing New Zealand - IVCNZ '12, 2012

ABSTRACT Smart cameras are extensively used for multi-view capture and 3D rendering applications. To achieve high quality, such applications are required to estimate accurate position and orientation of the cameras (called as camera calibration-pose estimation). Traditional techniques that use checkerboard or special markers, are impractical in larger spaces. Hence, feature-based calibration (auto-calibration), is necessary. Such calibration methods are carried out based on features extracted and matched between stereo pairs or multiple cameras. Well known feature extraction methods such as SIFT (Scale Invariant Feature Transform), SURF (Speeded-Up Robust Features) and ORB (Oriented FAST and Rotated BRIEF) have been used for auto-calibration. The accuracy of auto-calibration is sensitive to the accuracy of features extracted and matched between a stereo pair or multiple cameras. In practical imaging systems, we encounter several issues such as blur, lens distortion and thermal noise that affect the accuracy of feature detectors. In our study, we investigate the behaviour of SIFT, SURF and ORB through simulations of practical issues and evaluate their performance targeting 3D reconstruction (based on epipolar geometry of a stereo pair). Our experiments are carried out on two real-world stereo image datasets of various resolutions. Our experimental results show significant performance differences between feature extractors' performance in terms of accuracy, execution time and robustness to blur, lens distortion and thermal noise of various levels. Eventually, our study identifies suitable operating ranges that helps other researchers and developers of practical imaging solutions.

2012

Standard RANSAC does not perform very well for contaminated sets, when there is a majority of outliers. We present a method that overcomes this problem by transforming the problem into a 2D position vector space, where an ordinary cluster algorithm can be used to find a set of putative inliers. This set can then easily be handled by a modified version of RANSAC that draws samples from this set only and scores using the entire set. This approach works well for moderate differences in scale and rotation. For contaminated sets the increase in performance is in several orders of magnitude. We present results from testing the algorithm using the Direct Linear Transformation on aerial images and photographs used for panographs.

2013

Post-processing after matching technique such as RANSAC is useful for reducing outliners. However, such methods may not be able to increase the number of correctly matched pairs that is important in some application such as image stitching. In this work, a post-processing technique for increasing the number of correct matched points between two images with on-plane rotation is proposed. The proposed method makes use of the dominant rotational degree between two corresponding images to increase the number of matched points after feature extraction process using state-of-the-art methods such as Scale Invariant Feature Transform (SIFT) or Speeded-Up Robust Feature (SURF). The proposed method can generally increase the number of matched points around 10% to 20%. Furthermore, it can also correct the false matching which is caused by similar appearance of features.

International Journal of Innovative Research in Computer and Communication Engineering, 2013

Accurate, robust and automatic image registration is critical task in many applications. To perform image registration/alignment, required steps are: Feature detection, Feature matching, derivation of transformation function based on corresponding features in images and reconstruction of images based on derived transformation function. Accuracy of registered image depends on accurate feature detection and matching. So these two intermediate steps are very important in many image applications: image registration, computer vision, image mosaic etc. This paper presents two different methods for scale and rotation invariant interest point/feature detector and descriptor: Scale Invariant Feature Transform (SIFT) and Speed Up Robust Features (SURF). It also presents a way to extract distinctive invariant features from images that can be used to perform reliable matching between different views of an object/scene.

2005

This paper describes a novel multi-view matching framework based on a new type of invariant feature. Our features are located at Harris corners in discrete scale-space and oriented using a blurred local gradient. This defines a rotationally invariant frame in which we sample a feature descriptor, which consists of an 8 × 8 patch of bias/gain normalised intensity values. The density of features in the image is controlled using a novel adaptive non-maximal suppression algorithm, which gives a better spatial distribution of features than previous approaches. Matching is achieved using a fast nearest neighbour algorithm that indexes features based on their low frequency Haar wavelet coefficients. We also introduce a novel outlier rejection procedure that verifies a pairwise feature match based on a background distribution of incorrect feature matches. Feature matches are refined using RANSAC and used in an automatic 2D panorama stitcher that has been extensively tested on hundreds of sample inputs.

arXiv (Cornell University), 2022

This work proposes a new method for real-time dense 3d reconstruction for common 360°action cams, which can be mounted on small scouting UAVs during USAR missions. The proposed method extends a feature based Visual monocular SLAM (OpenVSLAM, based on the popular ORB-SLAM) for robust long-term localization on equirectangular video input by adding an additional densification thread that computes dense correspondences for any given keyframe with respect to a local keyframe-neighboorhood using a PatchMatch-Stereoapproach. While PatchMatch-Stereotypes of algorithms are considered state of the art for large scale Mutli-View-Stereo they had not been adapted so far for real-time dense 3d reconstruction tasks. This work describes a new massively parallel variant of the PatchMatch-Stereo-algorithm that differs from current approaches in two ways: First it supports the equirectangular camera model while other solutions are limited to the pinhole camera model. Second it is optimized for low latency while keeping a high level of completeness and accuracy. To achieve this it operates only on small sequences of keyframes, but employs techniques to compensate for the potential loss of accuracy due to the limited number of frames. Results demonstrate that dense 3d reconstruction is possible on a consumer grade laptop with a recent mobile GPU and that it is possible with improved accuracy and completeness over common offline-MVS solutions with comparable quality settings.

Procedings of the British Machine Vision Conference 2009, 2009

We analyze the usage of Speeded Up Robust Features (SURF) as local descriptors for face recognition. The effect of different feature extraction and viewpoint consistency constrained matching approaches are analyzed. Furthermore, a RANSAC based outlier removal for system combination is proposed. The proposed approach allows to match faces under partial occlusions, and even if they are not perfectly aligned or illuminated.

In this literature survey encompasses a detailed annotation of the novel scale and rotation invariant detector and descriptor, called SURF. Initially explores the overview of the SURF approach which terribly describes how SURF approach proceeds practically. It utilizes the integral image, hessian matrix and haar wavelet responses to improve the performance in robust way. And further discusses about utilities of SURF in computer vision systems. There broadly illustrates the applications such as multiple pedestrian tracking, medical image registration, face recognition and 3D reconstruction pipeline. Afterward this study elaborates optimization of SURF via utilizing dierent methodologies such as fully ane invariant SURF, color histograms and ACO approach. Finally this survey explicates pros and cons of SURF against other similar methodologies which specically focus on SIFT, PCA-SIFT and GLOH. There in substance demonstrates contrast between illumination changes, distinctiveness, repeatability and ane transformation.

Extremely variant image pairs include distorted, deteriorated, and corrupted scenes that have experienced severe geometric, photometric, or non-geometric-non-photometric transformations with respect to their originals. Real world visual data can become extremely dusty, smoky, dark, noisy, motion-blurred, affine, JPEG compressed, occluded, shadowed, virtually invisible, etc. Therefore, matching of extremely variant scenes is an important problem and computer vision solutions must have the capability to yield robust results no matter how complex the visual input is. Similarly, there is a need to evaluate feature detectors for such complex conditions. With standard settings, feature detection, description, and matching algorithms typically fail to produce significant number of correct matches in these types of images. Though, if full potential of the algorithms is applied by using extremely low thresholds, very encouraging results are obtained. In this paper, potential of 14 feature detectors: SIFT, SURF, KAZE, AKAZE, ORB, BRISK, AGAST, FAST, MSER, MSD, GFTT, Harris Corner Detector based GFTT, Harris Laplace Detector, and CenSurE has been evaluated for matching 10 extremely variant image pairs. MSD detected more than 1 million keypoints in one of the images and SIFT exhibited a repeatability score of 99.76% for the extremely noisy image pair but failed to yield high quantity of correct matches. Rich information is presented in terms of feature quantity, total feature matches, correct matches, and repeatability scores. Moreover, computational costs of 25 diverse feature detectors are reported towards the end, which can be used as a benchmark for comparison studies.