Ekta Walia

Thyroid Nodule Classification in Ultrasound Images by Fine-Tuning Deep Convolutional Neural Network

With many thyroid nodules being incidentally detected, it is important to identify as many malignant nodules as possible while excluding those that are highly likely to be benign from fine needle aspiration (FNA) biopsies or surgeries. This paper presents a computer-aided diagnosis (CAD) system for classifying thyroid nodules in ultrasound images. We use deep learning approach to extract features from thyroid ultrasound images. Ultrasound images are pre-processed to calibrate their scale and remove the artifacts. A pre-trained GoogLeNet model is then fine-tuned using the pre-processed image samples which leads to superior feature extraction. The extracted features of the thyroid ultrasound images are sent to a Cost-sensitive Random Forest classifier to classify the images into “malignant” and “benign” cases. The experimental results show the proposed fine-tuned GoogLeNet model achieves excellent classification performance, attaining 98.29% classification accuracy, 99.10% sensitivity and 93.90% specificity for the images in an open access database (Pedraza et al. 16), while 96.34% classification accuracy, 86% sensitivity and 99% specificity for the images in our local health region database.

Boosting local texture descriptors with Log-Gabor filters response for improved image retrieval

In the recent past, many local texture descriptors have been proposed for the image retrieval task. In order to improve the image retrieval accuracy, quite a few of these descriptors have been implemented on Gabor filter response. However, the response of Log-Gabor filters has been proved to be better than Gabor filters with respect to their discrimination ability. In this paper, we present a framework for image retrieval that applies various local texture descriptors on Log-Gabor filters response. To evaluate the retrieval performance of the proposed framework, experiments have been conducted on standard Wang, VisTex and OT-Scene databases. Consistent improvement in the image retrieval accuracy demonstrates the effectiveness of this framework. Further, the experimental results show that the use of proposed framework with low-dimension texture descriptors such as Orthogonal Combination of Local Binary Pattern makes them a better choice over Local Binary Pattern and its high-dimensional variants when higher retrieval accuracy, small feature vector size and ease of computation is desired.

Color texture description with novel local binary patterns for effective image retrieval

Abstract We propose a novel local color texture descriptor called local binary pattern for color images (LBPC). The proposed descriptor uses a plane to threshold color pixels in the neighborhood of a local window into two categories. To boost the discriminative power of the proposed LBPC operator, local binary patterns of the hue component in the HSI color space, called the local binary pattern of the hue (LBPH) is derived. Further, LBPC, LBPH are fused to derive LBPC+LBPH which when combined with the color histogram (CH) of the hue component results in an effective image retrieval method LBPC+LBPH+CH. The uniform patterns of the two proposed descriptors ULBPC and ULBPH are combined to yield another low dimension local color descriptor ULBPC+ULBPH+CH which provides a good tradeoff between retrieval accuracy and speed. Detailed experiments conducted on Wang, Holidays, Corel-5K and Corel-10K datasets demonstrate that the proposed low dimension descriptors LBPC+LBPH+CH, and ULBPC+ULBPH+CH outperform the state-of-the-art color texture descriptors in terms of retrieval accuracy and speed.

Partition selection with sparse autoencoders for content based image classification

Managing colossal image datasets with large dimensional hand-crafted features is no more feasible in most of the cases. Content based image classification (CBIC) of these large image datasets calls for the need of dimensionality reduction of features extracted for the purpose. This paper identifies the escalating challenges in the discussed domain and introduces a technique of feature dimension reduction by means of identifying region of interest in a given image with the use of reconstruction errors computed by sparse autoencoders. The automated process identifies the significant regions in an image for feature extraction. It not only improves the dimension of useful features but also contributes to increased classification results compared to earlier approaches. The reduction in number of one kind of features easily makes space for the inclusion of other features whose fusion facilitates improved classification performance compared to individual feature extraction techniques. Two different datasets, i.e. Wang dataset and Corel 5K dataset have been used for the experiments. State-of-the-art classifiers, i.e. Support Vector Machine and Extreme Learning Machine are used for CBIC. The proposed techniques are evaluated and compared in the context of both the classifiers and analysis of results suggests the appropriateness of the proposed methods for real time applications.

Improving Archaeologists’ Online Archive Experiences Through User-Centred Design

Traditionally, the preservation of archaeological data has been limited by the cost of materials and the physical space required to store them, but for the last 20 years, increasing amounts of digital data have been generated and stored online. New techniques in digital photography and document scanning have dramatically increased the amount of data that can be retained in digital format, while at the same time reducing the physical cost of production and storage. Vast numbers of hand written notes, grey literature documents, images of assemblages, contexts, and artefacts have been made available online. However, accessing these repositories is not always straightforward. Superficial interaction design, sparsely populated metadata, and heterogeneous schemas may prevent users from working the data that they need within archaeological archives. In this article, we present the work of the Digging into Archaeological Data and Image Search Metadata project (DADAISM), a multidisciplinary project that draws together the work of researchers from the fields of archaeology, interaction design, image processing and text mining to create an interactive system that supports archaeologists in their tasks in online archives. By adopting a user-centred approach with techniques grounded in contextual design, we identified the phases of archaeologists work in online archives, which are distinctive to this user group. The insights from this work drove the design and evaluation of an interactive system that successfully integrates content-based image based retrieval and improved metadata searching to deliver a positive user experience when working with online archives.

Image-based search and retrieval for biface artefacts using features capturing archaeologically significant characteristics

Archaeologists are currently producing huge numbers of digitized photographs to record and preserve artefact finds. These images are used to identify and categorize artefacts and reason about connections between artefacts and perform outreach to the public. However, finding specific types of images within collections remains a major challenge. Often, the metadata associated with images is sparse or is inconsistent. This makes keyword-based exploratory search difficult, leaving researchers to rely on serendipity and slowing down the research process. We present an image-based retrieval system that addresses this problem for biface artefacts. In order to identify artefact characteristics that need to be captured by image features, we conducted a contextual inquiry study with experts in bifaces. We then devised several descriptors for matching images of bifaces with similar artefacts. We evaluated the performance of these descriptors using measures that specifically look at the differences between the sets of images returned by the search system using different descriptors. Through this nuanced approach, we have provided a comprehensive analysis of the strengths and weaknesses of the different descriptors and identified implications for design in the search systems for archaeology.

Fast frequency domain method to detect skew in a document image

In this paper, a new fast frequency domain method based on Discrete Wavelet Transform and Fast Fourier Transform has been implemented for the determination of the skew angle in a document image. Firstly, image size reduction is done by using two-dimensional Discrete Wavelet Transform and then skew angle is computed using Fast Fourier Transform. Skew angle error is almost negligible. The proposed method is experimented using a large number of documents having skew between -90° and +90° and results are compared with Moments with Discrete Wavelet Transform method and other commonly used existing methods. It has been determined that this method works more efficiently than the existing methods. Also, it works with typed, picture documents having different fonts and resolutions. It overcomes the drawback of the recently proposed method of Moments with Discrete Wavelet Transform that does not work with picture documents.