Le Dong

Noise Estimation Using Mean Square Cross Prediction Error for Speech Enhancement

This paper shows the feasibility of noise extraction from noisy speech and presents a two-stage approach for speech enhancement. The preproposed mean square cross prediction error (MSCPE) based blind source extraction algorithm is utilized to extract the additive noise from the noisy speech signal in the first stage. After that a modified spectral subtraction and a modified Wiener filter approach are proposed to extract the speech signal for speech enhancement in the second stage, where all the frequency spectra of the extracted noise are utilized. Theoretical justification shows that the MSCPE-based algorithm can extract desired signal from mixed sources. Experimental results show that the averaged correlation coefficient between the extracted noise and the original additive noise are beyond 85% for Gaussian noise and beyond 75% for real-world noise at SNR = 0 dB, and the proposed speech enhancement approaches perform better than conventional methods, such as spectral subtraction and Wiener filter.

LSI: Latent semantic inference for natural image segmentation

Abstract We propose a novel label inference approach for segmenting natural images into perceptually meaningful regions. Each pixel is assigned a serial label indicating its category using a Markov Random Field (MRF) model. To this end, we introduce a framework for latent semantic inference of serial labels, called LSI, by integrating local pixel, global region, and scale information of an natural image into a MRF-inspired model. The key difference from traditional MRF based image segmentation methods is that we infer semantic segments in the label space instead of the pixel space. We first design a serial label formation algorithm named Color and Location Density Clustering (CLDC) to capture the local pixel information. Then we propose a label merging strategy to combine global cues of labels in the Cross-Region potential to grasp the contextual information within an image. In addition, to align with the structure of segmentation, a hierarchical label alignment mechanism is designed to formulate the Cross-Scale potential by utilizing the scale information to catch the hierarchy of image at different scales for final segmentation optimization. We evaluate the performance of the proposed approach on the Berkeley Segmentation Dataset and preferable results are achieved.

Scene-Oriented Hierarchical Classification of Blurry and Noisy Images

A system for scene-oriented hierarchical classification of blurry and noisy images is proposed. It attempts to simulate important features of the human visual perception. The underlying approach is based on three strategies: extraction of essential signatures captured from a global context, simulating the global pathway; highlight detection based on local conspicuous features of the reconstructed image, simulating the local pathway; and hierarchical classification of extracted features using probabilistic techniques. The techniques involved in hierarchical classification use input from both the local and global pathways. Visual context is exploited by a combination of Gabor filtering with the principal component analysis. In parallel, a pseudo-restoration process is applied together with an affine invariant approach to improve the accuracy in the detection of local conspicuous features. Subsequently, the local conspicuous features and the global essential signature are combined and clustered by a Monte Carlo approach. Finally, clustered features are fed to a self-organizing tree algorithm to generate the final hierarchical classification results. Selected representative results of a comprehensive experimental evaluation validate the proposed system.

Holons Visual Representation for Image Retrieval

Along with the enlargement of image scale, convolutional local features, such as SIFT, are ineffective for representing or indexing and more compact visual representations are required. Due to the intrinsic mechanism, the state-of-the-art vector of locally aggregated descriptors (VLAD) has a few limits. Based on this, we propose a new descriptor named holons visual representation (HVR). The proposed HVR is a derivative mutational self-contained combination of global and local information. It exploits both global characteristics and the statistic information of local descriptors in the image dataset. It also takes advantages of local features of each image and computes their distribution with respect to the entire local descriptor space. Accordingly, the HVR is computed by a two-layer hierarchical scheme, which splits the local feature space and obtains raw partitions, as well as the corresponding refined partitions. Then, according to the distances from the centroids of partition spaces to local features and their spatial correlation, we assign the local features into their nearest raw partitions and refined partitions to obtain the global description of an image. Compared with VLAD, HVR holds critical structure information and enhances the discriminative power of individual representation with a small amount of computation cost, while using the same memory overhead. Extensive experiments on several benchmark datasets demonstrate that the proposed HVR outperforms conventional approaches in terms of scalability as well as retrieval accuracy for images with similar intra local information.

Compact feature based clustering for large-scale image retrieval

This paper addresses the problem of fast similar image retrieval, especially for large-scale datasets with millions of images. We present a new framework which consists of two dependent algorithms. First, a new feature is proposed to represent images, which is dubbed compact feature based clustering (CFC). For each image, we first extract cluster centers of local features, and then calculate distribution histograms of local features and statistics of spatial information in each cluster to form compact features based clustering, replacing thousands of local features. It can reduce feature vectors of image representation and enhance the discriminative power of each feature. In addition, an efficient retrieval method is proposed, based on vocabulary tree through compact features based clustering. Extensive experiments on the Ukbench, Holidays, and ImageNet databases demonstrate that our method reduces the memory and computation overhead and improves the retrieval efficiency, while keeping approximate state-of-the-art accuracy.

A Hierarchical Distributed Processing Framework for Big Image Data

This paper introduces an effective processing framework nominated Image Cloud Processing (ICP) to powerfully cope with the data explosion in image processing field. While most previous researches focus on optimizing the image processing algorithms to gain higher efficiency, our work dedicates to providing a general framework for those image processing algorithms, which can be implemented in parallel so as to achieve a boost in time efficiency without compromising the results performance along with the increasing image scale. The proposed ICP framework consists of two mechanisms, i.e., Static ICP (SICP) and Dynamic ICP (DICP). Specifically, SICP is aimed at processing the big image data pre-stored in the distributed system, while DICP is proposed for dynamic input. To accomplish SICP, two novel data representations named P-Image and Big-Image are designed to cooperate with MapReduce to achieve more optimized configuration and higher efficiency. DICP is implemented through a parallel processing procedure working with the traditional processing mechanism of the distributed system. Representative results of comprehensive experiments on the challenging ImageNet dataset are selected to validate the capacity of our proposed ICP framework over the traditional state-of-the-art methods, both in time efficiency and quality of results.

Optimal kernel choice for domain adaption learning

In this paper, a kernel choice method is proposed for domain adaption, referred to as Optimal Kernel Choice Domain Adaption (OKCDA). It learns a robust classier and parameters associated with Multiple Kernel Learning side by side. Domain adaption kernel-based learning strategy has shown outstanding performance. It embeds two domains of different distributions, namely, the auxiliary and the target domains, into Hilbert Space, and exploits the labeled data from the source domain to train a robust kernel-based SVM classier for the target domain. We reduce the distributions mismatch by setting up a test statistic between the two domains based on the Maximum Mean Discrepancy (MMD) algorithm and minimize the Type II error, given an upper bound on error I. Simultaneously, we minimize the structural risk functional. In order to highlight the advantages of the proposed method, we tackle a text classification problem on 20 Newsgroups dataset and Email Spam dataset. The results demonstrate that our method exhibits outstanding performance. HighlightsWe propose a kernel choice method for domain adaption.We reduce the distribution mismatch based on the Maximum Mean Discrepancy.Given an upper bound on Type I error, our method minimizes the Type II error.We apply our method to classification and evaluate on two datasets.

Global-to-Local Oriented Rapid Scene Perception

An approach for rapid scene perception from global layout to local features is presented. The representation of a complex scene is initially built from a collection of global features from which properties related to the spatial layout of the scene and its semantic category can be estimated. The rapid perception of natural scenes relies partly on a global estimation of the features contained in the scene. Further analysis on the local essential areas is deployed on the basis. Such kind of integrated model guarantees the interactive processing between local and global features, thus enabling low-level features to initiate scene perception and categorization efficiently.

Multi-panel medical image segmentation framework for image retrieval system

The automatic segmentation of multi-panel medical images into sub-images improves the retrieval accuracy of medical image retrieval systems. However, the accuracy and efficiency of the available multi-panel medical image segmentation techniques are not satisfactory for multi-panel images containing homogenous color inter-panel borders and image boundary, heterogeneous color inter-panel borders, small size sub-images, or numerous number of sub-images. In order to improve the accuracy and efficiency, a Multi-panel Medical Image Segmentation Framework (MIS-Framework) is proposed and implemented based on locating the longest inter-panel border inside the boundary of the input image. We evaluated the proposed framework on a subset of imageCLEF 2013 dataset containing 2407 images. The proposed framework showed promising experimental results in terms of accuracy and efficiency on single panel as well as multi-panel image class identification and on sub-image separation as compared to the available techniques.

A color image retrieval system based on weighted average

Content-based image retrieval system needs a feature vector with dimensionality as lower as possible. In this paper, we propose an image retrieval system using low-dimensional color feature vector containing only one image feature termed as weighted average of colorful image. For finding similar images, the euclidean distance between the feature vector of query image and each feature vector of database images is calculated. We compared the proposed system with color averaging system presented in the literature on testing dataset and pascal dataset. The experiments demonstrate that our proposed system generates results more efficiently than the color averaging system without sacrificing the accuracy.