Ruth M. Aguilar-Ponce

Phase correlation based image alignment with subpixel accuracy

The phase correlation method is a well-known image alignment technique with broad applications in medical image processing, image stitching, and computer vision. This method relies on estimating the maximum of the phase-only correlation (POC) function, which is defined as the inverse Fourier transform of the normalized cross-spectrum between two images. The coordinates of the maximum correspond to the translation between the two images. One of the main drawbacks of this method, in its basic form, is that the location of the maximum can only be obtained with integer accuracy. In this paper, we propose a new technique to estimate the location with subpixel accuracy, by minimizing the magnitude of gradient of the POC function around a point near the maximum. We also present some experimental results where the proposed method shows an increased accuracy of at least one order of magnitude with respect to the base method. Finally, we illustrate the application of the proposed algorithm to the rigid registration of digital images.

Phase correlation with sub-pixel accuracy

Six methods for the accurate estimation of phase-correlation maxima are evaluated.Methods are tested under noise, extreme transformations, incomplete data, and for real cases with unknown transformations.Sinc function fitting provides the best average accuracy.Local Center of Mass, and Minimization of the POC gradient provide good balance between accuracy and efficiency. Six methods for the accurate estimation of the phase-correlation maxima are discussed and evaluated in this article for one- and two-dimensional signals. The evaluation was carried out under a rigid image registration framework, where artificially generated transformations were used in order to perform a quantitative assessment of the accuracy of each method and its robustness in the presence of noise, incomplete data, or extreme transformations. Another round of tests were performed with real cases where the true transformation is unknown, and not necessarily rigid; for these tests, quantitative evaluation was achieved by means of the root mean square error of the overlapping area between the two aligned images. While most methods behaved similarly under difficult conditions, three of the methods under study displayed clear advantages under mild levels of noise, low transformation complexity, and small percentages of missing data. These methods are the local center of mass, sinc function fitting, and minimization of the POC gradient magnitude. The other tested methods included quadratic fitting, linear fitting in the frequency domain, and up-sampling; however, these methods did not perform consistently well.

Phase-correlation guided area matching for realtime vision and video encoding

In computer vision and video encoding applications, one of the first and most important steps is to establish a pixel-to-pixel correspondence between two images of the same scene obtained at slightly different times or points of view. One of the most popular methods to find these correspondences, known as Area Matching, consists in performing a computationally intensive search for each pixel in the first image, around a neighborhood of the same pixel in the second image. In this work we propose a method which significantly reduces the search space to only a few candidates, and permits the implementation of real-time vision and video encoding algorithms which do not require specialized hardware such as GPU’s or FPGA’s. Theoretical and experimental support for this method is provided. Specifically, we present results from the application of the method to the realtime video compression and transmission, as well as the realtime estimation of dense optical flow and stereo disparity maps, where a basic implementation achieves up to 100 fps in a typical dual-core PC.

Background Subtraction based on Mutual Information

Motion Detection is major task in every application of computer vision such as video surveillance. Background Subtraction (BS) algorithms have been employed for several years to find a moving objects in a scene. BS has been used in video surveillance due to its simplicity and the fact that cameras are stationary in a video surveillance systems. The present paper introduce a new approach to motion detection through a hierarchical model that uses Mutual Information as a measure of change in the scene. Since pixels in a frame belong to objects, a segmentation in regions of the frame is done by mean-shift algorithm. Then a Mutual information measurement between the segmented region and the incoming frame is performed. A first approach to foreground mask is achieved and later is refined using a modification of the Wronskian Change Detector (WCD). The experimental results show that our proposed algorithm improve the performance in comparison with a pixel based Background Subtraction algorithm mixture of gaussians (MoG), a hierarchical block based Background Subtraction algorithm (HMDRP) and a test of linear independence (WCD).

POC-guided are matching algorithm for video encoding

A motion estimation algorithm based on Phase-Only Correlation (POC) function is proposed. Motion estimation is one of the key tasks in several applications such as video encoding. Motion estimation is the process of finding the vertical and horizontal displacement of a block. The accuracy of this process depends on the search method and matching criteria. The proposed method calculates POC on the entire frame.POC gives us a good estimated of the displacement that occurs within the frame. The size of the peaks depends on the size of the area that is moving towards the direction represented by them. We select a set of peaks that can be coded in a lookup table. Then, Sum of Absolute Difference (SAD) is calculated for each vector within the table. The vector with the lowest SAD for the block is selected as the motion vector. The search method based on POC provides better results compared with Full Search.

Detection and classification of intra boby acoustic wave signals for wearable applications

Wearable Devices (WD) are systems designed to do a specific task, these system are embedded in daily life personal objects. Usual transducers in wearable devices involve accelerometers gyroscopes, cameras etc. In WD is required to design efficiently in terms of power. Therefore, a new trend incorporates acoustic transducer that does not need a power supply to sense. They are very cheap and easy to replace. In WD, these transducers detect acoustic wave signals traveling in human tissue and bones. The processing of these kind of signals is related with many areas such as medical, gesture recognition, haptics etc. The proposed system uses a set of three sensors to capture Intra Body Acoustic Wave Signals (IBAWS) from the wrist of the right hand. A signal database is constructed using 18 users repeating 30 times each one of five gestures proposed. The patterns are composed by six features per sensor, including Spectral Flux, Spectral Centroid and Short Time Energy. Complete proposed patterns contains 18 features. Classifiers results shown 75.37% of accuracy using Bayesian classifier with Gaussian Kernel, 80% of accuracy using Knn classifier, and 85.56% of accuracy with Artificial Neural Networks. All this for a set of 18 users, supporting the hypothesis that classify IBAWS is independent from the user and could be generalized to use them in WD.

Architecture for Real-Time Color Detection on Video Based on the Choquet Fuzzy Integral

Segmentation is the processes of classification of pixels according to certain criteria. Segmentation is widely used in a variety of application including monitoring, surveillance and object detection. The Choquet Fuzzy Integral has been used for segmentation achieving good results. The present work proposes an architecture for Choquet Fuzzy Integral implemented in VHDL. The current implementation achieves real-time execution.

Four Level Haar Transform Architecture for Feature Extraction

Biomedical Electronic Devices have been developed as aid in some neurological conditions like epilepsy and Parkinson. Those devices require low-power systems in order to guarantee portability and continuous operation from weeks to years. Therefore, the present work proposes a simplified architecture for Haar Wavelet Transform used as feature extraction in the spike sorting process. The architecture achieves a reduction of 46% on the number of multipliers needed in a direct architecture. As result of the reduction onmultiplication, the rounding off error in the proposed architecture is also reduced achieving zero error in spike sorting classification.

Estimation of gait stability based on accelerometer signals

An estimation of gait stability based on accelerometer signal is presented. The walking process of a human body is a complex task that involves the muscular, joint and nervous systems. The gait cycle is divided into swing and stance phases. Gait stability is defined as repetition of the gait cycle with minimum variability. In order to assets the gait stability a series of parameter are estimated, such parameters include, duration of the gait cycle, length of the left and right step, average speed of the gait and number of samples per cycle. The first step toward establish gait stability is estimated the beginning and ending of a cycle. This process is achieved by a series of filter that isolated 2Hz component that is produce by heel strike. Once the cycles have been localized, the rest of the parameters are estimated. The algorithm was tested on a set of 40 healthy persons from ages 20 to 59. The results show that the stability decreases with age, as expected, since our neuromotor system deteriorates with age.

Autonomous Decentralized Systems Based Approach to Object Detection in Sensor Clusters

This work deploys Autonomous Decentralized System (ADS) based formulation to cluster of networked visual sensors. The goal is to utilize and integrate the sensing and networking capabilities of the sensors with the systematic and autonomous features of ADS to perform visual surveillance through object detection in the covered areas of interest. In the proposed approach, several cells are distributed through an area of interest called Autonomous Observer Cell. The decentralized subsystems detect and track moving objects present on the scene by looking through a camera embedded in each sensor. These subsystems form a cluster and each cluster sends information to an Autonomous Analysis Cell that determines if an object of interest is present. The Autonomous Observer Cells share a common data field and a cluster-head works as a gateway between the cluster and the Autonomous Analysis Cell.