Harikrishna G. N. Rai

Hybrid feature to encode shape and texture for Content Based Image Retrieval

In this paper, we propose an approach for representing both shape and texture information in an image using a single hybrid feature descriptor for Content Based Image Retrieval. Towards this, we compute the gradient magnitude of the input image prior to deriving features. Feature extraction is then performed using the responses from a bank of Gabor filters. Here, we exploit the fact that shape corresponds to the high spatial frequency content in the image whereas natural texture information predominantly lies within low to mid-range frequencies. This approach helps in better localization of characteristic texture as well as shape, due to spread of energy towards high frequencies in spectral domain. Moment invariants are extracted from Gabor filter responses which yield better retrieval performance than conventional statistical features. Experimental results show that this approach has relatively improved retrieval performance on Corel image data set when compared with recent approaches in the literature. Further experiments were also performed on a medical image dataset with 95.4 percent precision and 74.6 percent recall.

Video analytics solution for tracking customer locations in retail shopping malls

Due to increased adoption of digital inclusion in various businesses, location based services are gaining importance to provide value-added services for their customers. In this work, we present a computer vision based system for tracking customer locations by recognizing individual shopping carts inside shopping malls in order to facilitate location based services. We provide an efficient approach for cart recognition that consists of two stages: cart detection and then cart recognition. A binary pattern is placed between two pre-defined color markers and attached to each cart for recognition. The system takes live video feed as input from the cameras mounted on the aisles of the shopping mall and processes frames in real-time. In the cart detection stage, color segmentation, feature extraction and classification are used for detection of binary pattern along with color markers. In recognition stage, segmented binary strip is processed using spatial image processing techniques to decode the cart identification number.

An efficient shape based feature for retrieval of healthcare literatures using CBIR technique

Recent advances in healthcare such as Evidence Based Medicine (EBM) and Clinical Decision Support Systems (CDSS) requires practitioners to frequently access archived historical healthcare literatures and images. As the majority of healthcare literatures contain images such as medical images, clip arts, waveforms, flow charts and block diagrams, in this paper we present the use of Content Based Image Retrieval (CBIR) for efficient healthcare literature search and retrieval. We introduce a novel shape based feature called Fourier Edge Orientation Autocorrelogram (FEOAC) for search and retrieval of healthcare literatures. Scale and translation invariant Edge Orientation Autocorrelogram (EOAC) feature is made rotation invariant by applying Fourier transform. This Fourier based shape feature also reduces the feature set dimension enabling faster retrieval of document images in large databases. Experimental results show that FEOAC outperforms EOAC for search and retrieval of healthcare document images, with improved precision and recall rates.

An Innovative System for Remote and Automated Testing of Mobile Phone Applications

Application testing is an integral part of the application life cycle. This testing effort is more for the 3rd party applications in the mobile phone market, due to the wide number of handsets available on which the application needs to be tested before being released. At the same time, majority of the applications use the cellular network, necessitating the tester1 (along with the handset) to be present in the service area of the cellular network. We present a remote testing system, wherein the handset is in the cellular network service area, but the tester is present in a remote location. The tester controls the handset over the Internet. This gives opportunities to leverage the potential of the global outsourcing business model in mobile application testing domain. In addition, the system is agnostic to the Operating System & application running on the mobile phone, and is also non-intrusive. Further, we present preliminary results on automating this remote testing process itself.

Automatic blur detection in mobile captured document images: Towards quality check in mobile based document imaging applications

Optical Character Recognition is widely used for automated processing of document images. While character recognition technology is mature, its application to mobile captured document image is still at its nascent stage. Capturing images from a mobile camera poses several challenges like motion blur, defocus and geometrical distortions which are usually not encountered in scanned or calibrated camera captured images. Therefore determining the quality of images automatically prior to recognition is an important problem. Quality check is especially useful in financial transaction instruments like bill payment where accuracy of text recognition for sensitive fields such as “amount due” should be high. Poor quality images can be rejected prior to OCR to avoid incorrect text recognition and save processing time. This paper discusses some techniques in literature for blur detection in mobile camera captured document images. We propose a simple yet elegant method that addresses some challenges faced in these document images. Extensive testing is performed on large dataset containing more than 4000 mobile captured images and optimum parameter values for performing quality check against motion blur and defocus are identified. Our experimental results demonstrate the effectiveness of the proposed method. In addition we realized a smart mobile application for blur detection and report its performance on several mobile devices.

A Smart Mobile Application for Identifying Storage Location of Small Industrial Assets

Automation of industrial asset management has gained significant attention due to recent breakthrough in technologies for object identification and tracking. Unavailability of small assets used in industrial assembly floors, owing to post use misplacement, impacts the productivity and results in erroneous information in inventory management systems. In this work we present a solution for identifying small industrial assets along with their storage location using a machine vision based approach. Proposed system will assist the store manager to track and consolidate industrial assets periodically on the assembly floor after production. We have developed a machine vision based application on a mobile handheld device for asset tracking. This application is capable of acquiring asset images and finding their storage location. A mobile application prototype is implemented on an open software platform for evaluation.

Figure Retrieval in Biomedical Literature

Automatic classification of figures present in healthcare documents is known to be useful for biomedical document mining. The context of a document is directly reflected in the figures present within them. Embedded text within these figures along with image features have been used for figure retrieval. We demonstrate that image features based on structural properties of figures alone is sufficient for the figure retrieval task. An algorithm for describing structural properties of the embedded images, Fourier Edge Orientation Autocorrelogram, which utilizes spatial distribution of detected edges, is presented. We have shown that Fourier Edge Orientation Autocorrelogram performs better than its predecessor, when most of the edge information is retained. The algorithm is validated on publicly available figures from healthcare literature. Apart from invariance to scale, rotation and non-uniform illumination, the proposed feature descriptor is also shown to be relatively robust to noisy edges. Since there is no standard dataset available for figure classification, comparison of the proposed feature descriptor with four well known binary shape descriptors is demonstrated. The retrieval performance shows an overall improvement over other known methods in figure retrieval task.

Texture edge statistics for efficient retrieval of biomedical images

Adoption of Clinical Decision Support Systems in the process of clinical decision process has been gaining attention in recent times. Such intelligent decision support systems need frequent access to historic medical data such as medical images and associated reports. Content Based Image Retrieval has been preferred choice of technique for such smart retrieval of medical images based on their content. In this paper, we introduce a new approach for deriving edge based features for retrieval of medical images. Texture edges are known to represent the tissue boundaries better than intensity edges in medical images. Therefore, we use texture edges for describing image structure instead of traditional approach of using intensity edges. We demonstrate that retrieval performance at organ level in medical images using texture edges is superior to intensity edges.

Figure Based Biomedical Document Retrieval System using Structural Image Features

Multi-modal and Unstructured nature of documents make their retrieval from healthcare document repositories a challenging task. Text based retrieval is the conventional approach used for solving this problem. In this paper, the authors explore an alternate avenue of using embedded figures for the retrieval task. Usually, context of a document is directly reflected in the associated figures, therefore embedded text within these figures along with image features have been used for similarity based retrieval of figures. The present work demonstrates that image features describing the structural properties of figures are sufficient for the figure retrieval task. First, the authors analyze the problem of figure retrieval from biomedical literature and identify significant classes of figures. Second, they use edge information as a means to discriminate between structural properties of each figure category. Finally, the authors present a methodology using a novel feature descriptor namely Fourier Edge Orientation Autocorrelogram FEOAC to describe structural properties of figures and build an effective Biomedical document retrieval system. The experimental results demonstrate the better retrieval performance and overall improvement of FEOAC for figure retrieval task, especially when most of the edge information is retained. Apart from invariance to scale, rotation and non-uniform illumination, the proposed feature descriptor is shown to be relatively robust to noisy edges.