Search

Date of Conference: 16-19 May 2016 Conference Name: IEEE 24th Signal Processing and Communications Applications Conference, SIU 2016 In this paper, we study multi-class classification of tweets, where we introduce highly efficient dimensionality reduction techniques suitable for online processing of high dimensional feature vectors generated from freely-worded text. As for the real life case study, we work on tweets in the Turkish language, however, our methods are generic and can be used for other languages as clearly explained in the paper. Since we work on a real life application and the tweets are freely worded, we introduce text correction, normalization and root finding algorithms. Although text processing and classification are highly important due to many applications such as emotion recognition, advertisement selection, etc., online classification and regression algorithms over text are limited due to need for high dimensional vectors to represent natural text inputs. We overcome such limitations by showing that randomized projections and piecewise linear models can be efficiently leveraged to significantly reduce the computational cost for feature vector extraction from the tweets. Hence, we can perform multi-class tweet classification and regression in real time. We demonstrate our results over tweets collected from a real life case study where the tweets are freely-worded, e.g., with emoticons, shortened words, special characters, etc., and are unstructured. We implement several well-known machine learning algorithms as well as novel regression methods and demonstrate that we can significantly reduce the computational complexity with insignificant change in the classification and regression performance.

Date of Conference: 2-5 May 2018 In this paper, we study online nonlinear data regression and propose a highly efficient long short term memory (LSTM) network based architecture. Here, we also introduce on-line training algorithms to learn the parameters of the introduced architecture. We first propose an LSTM based architecture for data regression. To diminish the complexity of this architecture, we use an energy efficient operator (ef-operator) instead of the multiplication operation. We then factorize the matrices of the LSTM network to reduce the total number of parameters to be learned. In order to train the parameters of this structure, we introduce online learning methods based on the exponentiated gradient (EG) and stochastic gradient descent (SGD) algorithms. Experimental results demonstrate considerable performance and efficiency improvements provided by the introduced architecture.

Compressed sensing theory shows that any signal which is defined as sparse in a given domain can be reconstructed using fewer linear projections instead of using all Nyquist-rate samples. In this paper, we investigate basis pursuit, matching pursuit, orthogonal matching pursuit and compressive sampling matching pursuit algorithms, which are basic compressed sensing based algorithms, and present performance curves in terms of mean squared error for various parameters including signal-to-noise ratio, sparsity and number of measurements with regard to mean squared error. In addition, accuracy of estimation performances has been supported with theoretical lower bounds (Cramer-Rao lower bound and deterministic lower mean squared error). Considering estimation performances, compressive sampling matching pursuit yields the best results unless the signal has a non-sparse structure.

Date of Conference: 20-22 April 2011 Conference Name: 19th Signal Processing and Communications Applications Conference, SIU 2011 In this paper, a new histogram based auto-focusing method for thermal cameras is proposed. This proposed method is realized by FPGA (Field Programmable Gate Array) and DSP (Digital Signal Processor) working together and simultaneously. HF (High Frequency) component, obtained from real-time image flow by FPGA and DSP is used for auto-focusing process. Proposed method is able to determine the focus direction from the HF component produced in the process of histogram equalization by FPGA, unlike Fourier transform and pixel differenve based methods in the literature. With this superiority, proposed method requires no extra calculation for thermal cameras for which histogram equalization is necessary. Analysis show that proposed method is successful on the simulations and scanning thermal cameras.

A powerful method in analysis of large data sets where there are many natural clusters with varying statistics such as different sizes, shapes, density distribution, is the use of sel-forganizing maps (SOMs) [1]. However, further processing tools, such as visualization, interactive clustering, are often necessary to capture the clusters from the learned SOM knowledge. A recent visualization scheme, CONNvis [2], and interactive clustering from CONNvis, utilizes the data topology for SOM knowledge representation by using a weighted Delaunay graph, CONN. In this paper, an automated clustering scheme for SOMs, SOMcluster, which is a two-level clustering of CONN by the skills obtained in the interactive process, is proposed. It is shown that SOMcluster, which does not require the number of clusters a priori, is used successfully for automated segmentation of a remote sensing spectral image which has many clusters some of which were unidentified in previous works.

In stereoscopic 3D video, it is well-known that humans can perceive high quality 3D video provided that one of the views is in high quality. Hence, in stereo video encoding, the best overall rate vs. perceived-distortion performance may be achieved by reduction of the spatial, temporal and/or quantization resolution of the second view, while keeping the first view in full resolution. Objective stereo video quality metric is a crucial point for the success of unequal rate allocation. Since the perceived 3D video quality does not correlate well with the average PSNR of the two views, we proposed a new quantitative measure using a weighted combination of two PSNR values and a jerkiness measure [1]. In this paper, we investigate whether the PSNR weights should be related with spatial activity in different scaling options and video contents. Spatial activity is examined in terms of blockiness and blur. The optimal weights for some specific experiments are determined by matching, i.e. searching the best correlation between, subjective quality test results and the objective measure scores on a set of test videos.

Date of Conference: 9-11 April 2009 Conference Name: IEEE 17th Signal Processing and Communications Applications Conference, SIU 2009 In the last few years, there is great increase in capture and representation of real 3-Dimensonal scenes using 3D animation models. The 3D signals are then compressed, transmitted to the client side and reconstructed for the user view. Each step mentioned here opened a new subject in the field of signal processing. While processing these models, using the model as a whole is not the best approach. Therefore clustering the model vertices became a very common method. For example, it is very common to use motion based clustering in animation compression. In this paper a new dynamic model clustering algorithm is proposed. Animation vertices are first put through PCA and partitioned into their eigenvalues and eigenvectors. The eigenvectors found using the proposed method can be called eigentrajectories. Then the dot product of the these eigentrajectories with the trajectories of the animation vertice are found. These coefficients are used to cluster the animation model. The results and the comparisons with a similar approach show that the proposed algorithm is successful.