Shahedur Rahman

Parallel image processing system-a modular architecture using dedicated image processing modules and a graphics processor

The authors present the dedicated hardware features of an image processing system. The system is based on a modular architecture where all modules are independent of each other. A convolver implementation is given as an example of this approach. The hardware oriented image processing system is designed around a graphics processor. The system consists of a number of processing modules to enhance the performance of typical image processing applications in real-time and uses video RAM and dual-port RAM to enhance the system speed. The proposed design is a real-time image processing system. The novel feature of this architecture is that the algorithm used to generate the output data is programmable and uses separate dedicated image processing modules. These are combined to form a dedicated processing element (PE) which performs a range of tasks, eg, averaging, filtering, edge detection, scaling, point enhancement and thresholding.<<ETX>>

A novel low complexity Local Hybrid Pseudo-SSIM-SATD distortion metric towards perceptual rate control

The front-end block-based video encoder applies an Image Quality Assessment (IQA) as part of the distortion metric. Typically, the distortion metric applies uniform weighting for the absolute differences within a Sub-Macroblock (Sub-MB) at any given time. As video is predominately designed for Humans, the distortion metric should reflect the Human Visual System (HVS). Thus, a perceptual distortion metric (PDM), will lower the convex hull of the Rate-Distortion (R-D) curve towards the origin, by removing perceptual redundancy and retaining perceptual clues. Structured Similarity (SSIM), a perceptual IQA, has been adapted via logarithmic functions to measure distortion, however, it is restricted to the Group of Picture level and hence unable to adapt to the local Sub-MB changes. This paper proposes a Local Hybrid Pseudo-SSIM-SATD (LHPSS) Distortion Metric, operating at the Sub-MB level and satisfying the Triangle Equality Rule (≤). A detailed discussion of LHPSSs Psuedo-SSIM model will illustrate how SSIM can be perceptually scaled within the distortion metric space of SATD using non-logarithmic functions. Results of HD video encoded across different QPs will be presented showing the competitive bit usage under IbBbBbBbP prediction structure for similar image quality. Finally, the mode decision choices superimposed on the Intra frame will illustrate that LHPSS lowers the R-D curve as homogeneous regions are represented with larger block size.

Review of Standard Traditional Distortion Metrics and a need for Perceptual Distortion Metric at a (Sub) Macroblock Level

Within a video encoder the distortion metric performs an Image Quality Assessment (IQA). However, to exploit perceptual redundancy to lower the convex hull of the Rate-Distortion (R-D) curve, a Perceptual Distortion Metric (PDM) modelling of the Human Visual System (HVS) should be used. Since block-based video encoders like H.264/AVC operate at the Sub-Macroblock (Sub-MB) level, there exists a need to produce a locally operating PDM. A locally operating PDM must meet the requirements of Standard Traditional Distortion Metrics (STDMs), in that it must satisfy the Triangle Equality Rule (≤). Hence, this paper presents a review of STDMs of SSE, SAD and SATD against the perceptual IQA of Structural Similarity (SSIM) at the Sub-MB level. Furthermore, this paper illustrates the Universal Bounded Region (UBR) by block size that supports the triangle equality rule (≤) within the Sub-MB level, between SSIM and STDMs like SATD at the prediction stage.

An approach to develop human gene disorder database for intelligent variance analysis of genes and its products

Molecular biology databases presently exist in different formats, different sizes and in different data structures. These create a new challenge which is to make these databases interoperable. Currently, hyperlink text is attempting to link these databases but it is unable to resolve semantic conflicts of the data. Data interpretation through image visualization and smart query processing using object keying will ease the biological query many times. Our aim is to investigate and to develop a conceptual object model for a genetic disorder disease database which will interact with other heterogeneous public domain molecular biology databases for more meaningful information. The research focuses on issues such as data input, access, format, storage, searching using image key, semantics and interoperabilities when building such a specialized gene disorder database data model.

Instant Channel Allocation Technique to Improve System Throughput in Joint LTE Cellular Network

Femtocells can be easily deployed in macrocell LTE network without requiring a centralized planning to provide high data rate connectivity with a limited coverage. Femtocell is low-power, very small and cost effective cellular based station used in indoor environment. Femtocell and Macrocell share same frequency band which cause destructive interference. In order to maintain reliable service of macro-cells, it is most important to mitigate destructive femto to macro interference. Frequency Reuse (FR) is an effort of manipulating the frequency resource allocation upon terminals location to improve system throughput. In this paper, an efficient method to improve system capacity through interference management in the existing Femto-Macro two tier networks has been proposed. In the proposed scheme, femtocell user will be allocated frequency instantly as it is required. This will also improve system throughput which will effectively control interference between macrocells and femtocells.

Efficient and robust detection and recognition of objects in grayscale images

A new method for the detection and recognition of objects was developed for grayscale images. Obstacle detection is based on an efficient binarization and enhancement techniques followed by a suitable connected component analysis procedure. The grayscale object corresponding to the object identified in the binary image is then extracted. The second step deals with the recognition of these extracted objects. Each object is then described by Zernike moments. To achieve rotation and scaling invariance an efficient method based on bounding box is used. In order to achieve better results for object recognition, modified Support Vector Machine(SVM) classifiers utilizing decision tree for solving multiclass problems is used. The algorithm performs the task of object detection and recognition more efficiently, even with external constraints i.e. image scenes can have Shadows, partial occlusion and non-uniform illumination and at a much faster rate. The efficiency of the proposed method on grayscale images is shown by cascading some objects from COIL −8 database.

Integration of biological data resources using image object keying

This paper presents a novel approach to integrate biological data from multiple resources into a single page using image object keying. In this interactive approach, a gel electrophoresis protein spot is selected by the user which initiates the retrieval of corresponding 3D structure of the protein. It provides a set of operators to access and to collect elements content of the biological data resources for integration. The approach utilises a number of tools, namely, meta data extractor, mapping linker, dispatcher and result integrator for unifying the data collected from multiple data sources.

A modular based parallel image processing system

An image processing system which is designed around dedicated hardware devices is described. The system is based on a modular architecture where all modules are independent of each other. A convolver implementation is given as an example of this approach. The hardware-oriented image processing system is designed around a graphic processor. The system consists of a number of processing modules to enhance the performance of typical image processing applications in real-time and uses video RAM and dual-port RAM to enhance the system speed. The design aims to achieve real-time processing of images.<<ETX>>

Low complexity sub-block perceptual distortion assessment for mode decision and rate-control

Video is being used in a variety of portable and low powered devices, via online/on-demand services, for personal communications or over heterogeneous/wireless sensor networks. Likewise, perceptual evaluation is being sought, to ensure video sequences maintain perceptual integrity. This raises a challenge, to bring high complexity perceptual algorithms into a low complexity environment. Existing perceptual solutions minimise the overall complexity by making the Lagrange multiplier (λ), the quantisation stage perceptually aware. These solutions are restricted to the block level using the original pixels, a model or previous encoded block, thus avoiding assessing individual subblock candidates. Current perceptual algorithms like structured similarity (SSIM) uses statistical based calculations, and in order to be compatible with existing scores a further high complexity function is required for scaling. This paper presents a perceptual distortion and activity assessment that can operate at the subblock for each candidate during the later stages of encoding, in mode-decision and in rate-control, without the need for statistical calculations nor the high complexity associated with scaling a perceptual algorithm. The paper will show how several perceptual techniques of SSIM luma function, just noticeable detection (JND) and a new proposed edge detection can be used to form a low complexity solution. Consequently, the proposed low perceptual assessment has additional timing increase of <; +4 % for medium and low activity video sequences.

A framework for molecular biology database integration using context graph

This paper proposes a framework based on the context of the Web data for interactions with different biological data resources so that a consolidated view of data can be achieved. A formal description of context of resources and their relationships have been described here using context graph. The context increases the interoperability by providing the description of the resources and the navigation plan for accessing the Web based databases. A higher level construct is developed to implement the context in RDF for Web interactions. The interaction among the resources is achieved by using a context based integration domain. The integration domain allows to navigate and execute the query plan within the resource databases.