Nabeel Mohammed

Efficient and accurate independent component filter-based features for texure similarity

This paper evaluates the accuracy and efficiency of collection-specific texture features for Content-Based Image Retrieval. Independent Component Analysis is used to extract Independent Component Filters (ICF) from an image set. As these ICF are learned from the image set, the hypothesis is that they should provide texture features that are more effective than those extracted using generic filter banks. We describe a method for extracting candidate ICF from an image set, and choosing a representative subset from them. These are then used to extract image features. A simple CBIR system has been developed to evaluate the performance of these features on two standard texture image collections, compared with features extracted using multiple banks of Gabor filters. The results indicate that ICF-based features perform better than Gabor-based features, even when a much smaller number of ICF features is used than Gabor features. The ICF features are thus more accurate, and more efficient.

Effectiveness of ICF features for collection-specific CBIR

This study aims to find more effective methods for collection-specific CBIR. A lot of work has been done in trying to adapt a system by user feedback, in this study we aim to adapt CBIR systems for specific image collections in an automated manner. Independent Component Analysis (ICA), a high order statistical technique, is used to extract Independent Component Filters (ICF) from image sets. As these filters are adapted to the data, the hypothesis is that they may provide features which are more effective for collection-specific CBIR. To test this question, this study develops a methodology to extract ICF from image sets and use them to extract filter responses. In developing this method, the study uses image cross-correlation and clustering to solve issues to do with shifted/duplicate filters and selecting a smaller set of filters to make CBIR practical. The method is used to generate filter responses for the VisTex database . The filter response energies are used as features in the GNU Image Finding Tool (GIFT). The experiments show that features extracted using ICF have the potential to improve the effectiveness of collection-specific CBIR, although some more work in this area is required.

BanglaLekha-Isolated: A multi-purpose comprehensive dataset of Handwritten Bangla Isolated characters

BanglaLekha-Isolated, a Bangla handwritten isolated character dataset is presented in this article. This dataset contains 84 different characters comprising of 50 Bangla basic characters, 10 Bangla numerals and 24 selected compound characters. 2000 handwriting samples for each of the 84 characters were collected, digitized and pre-processed. After discarding mistakes and scribbles, 1,66,105 handwritten character images were included in the final dataset. The dataset also includes labels indicating the age and the gender of the subjects from whom the samples were collected. This dataset could be used not only for optical handwriting recognition research but also to explore the influence of gender and age on handwriting. The dataset is publicly available at https://data.mendeley.com/datasets/hf6sf8zrkc/2.

Bangla handwritten digit recognition using autoencoder and deep convolutional neural network

Handwritten digit recognition is a typical image classification problem. Convolutional neural networks, also known as ConvNets, are powerful classification models for such tasks. As different languages have different styles and shapes of their numeral digits, accuracy rates of the models vary from each other and from language to language. However, unsupervised pre-training in such situation has shown improved accuracy for classification tasks, though no such work has been found for Bangla digit recognition. This paper presents the use of unsupervised pre-training using autoencoder with deep ConvNet in order to recognize handwritten Bangla digits, i.e., 0–9. The datasets that are used in this paper are CMATERDB 3.1.1 and a dataset published by the Indian Statistical Institute (ISI). This paper studies four different combinations of these two datasets-two experiments are done against their own training and testing images, other two experiments are done cross validating the datasets. In one of these four experiments, the proposed approach achieves 99.50% accuracy, which is so far the best for recognizing handwritten Bangla digits. The ConvNet model is trained with 19,313 images of ISI handwritten character dataset and tested with images of CMATERDB dataset.

Classification of Bangla Compound Characters Using a HOG-CNN Hybrid Model

Automatic handwriting recognition is challenging task due to its sheer variety of acceptable stylistic differences. This is especially true for scripts with large character sets. Bangla, the sixth most widely spoken language in the world has a complex, large and rich set of compound characters. In this study, a hybrid deep learning model is proposed which combines the use of the manually designed feature Histogram of Oriented Gradients (HOG), with the adaptively learned features of a Convolutional Neural Networks (CNN). The proposed hybrid model was trained on the CMATERDB 3.1.3.3, a Bangla compound character data set which divides Bangla compound characters into 177 broad classes and 199 specific classes. The results demonstrate that CNN-only models achieve over 91% and 92% test accuracy respectively. Furthermore, it is shown that the proposed model, which incorporates HOG features with a CNN, achieves over 92.50% test accuracy on each division. While there is still room for improvement, these results are significantly better than currently published state of art on this data set.

Optimal range estimation for energy efficient dynamic packet size

In this paper energy efficient optimal packet size range is investigated for ad hoc networks. The energy efficiency of optimal packet size depends on a number of factors including channel noise, number of sources and hops, and many physical layer parameters (e.g. path loss, fading, transmitter/receiver power). Therefore, changing the packet size dynamically according to scenarios has been established as a convenient process. However, as there are numbers of parameters involved in size allocation, it is often difficult to fixing the packet size at runtime. Hence, this paper presents an energy efficient optimal packet size range. The observation shows that if that range is retained during the communication process, at least 10%–30% more energy efficiency can be achieved. Efficiency is investigated for time varying scenarios of ad hoc networks using Ns-2 simulator and compared with theoretical estimation. The simulation results matched with the theoretical estimation with low standard deviation.

Image augmentation by blocky artifact in Deep Convolutional Neural Network for handwritten digit recognition

Deep Convolutional Neural Networks - also known as DCNN - are powerful models for different visual pattern classification problems. Many works in this field use image augmentation at the training phase to achieve better accuracy. This paper presents blocky artifact as an augmentation technique to increase the accuracy of DCNN for handwritten digit recognition, both English and Bangla digits, i.e., 0–9. This paper conducts a number of experiments on three different datasets: MNIST Dataset, CMATERDB 3.1.1 Dataset and Indian Statistical Institute (ISI) Dataset. For each dataset, DCNNs with the proposed augmentation technique give better results than those without such augmentation. Unsupervised pre-training with the blocky artifact achieves 99.56%, 99.83% and 99.35% accuracy respectively on MNIST, CMATERDDB and ISI datasets producing, in the process, so far the best accuracy rate for CMATERDB and ISI datasets.

A packet level simulation study of adhoc network with network simulator-2 (NS-2)

Researchers in the field of computer networking often relies on appropriate simulators to diagnose network and to analyze the impact of various parameters on the network performance. There now exists several simulators (in the areas of wireless ad-hoc network) that offer various features to measure network performance. One such simulator that recently gained high popularity among researchers in the area of networking is Network Simulator 2 (NS-2). In this paper, we present a short survey on NS-2 for packet level simulation. The paper demonstrates how packet level simulation is conducted using NS-2. The paper furthermore explores the advantages and the drawbacks of NS-2 in the light of packet loss, interval period, traffic type and pattern. The techniques used in this work to measure network performance can be replicated in other related work. Simulated results with the NS-2 simulator show remarkable correlation with that of the theoretical estimation. The most remarkable contribution of this paper is this paper shows simulation for packet level analysis with NS-2 with details parameter setting, network scenarios and other necessary configurations which will lead the researchers to investigate network efficiency and performance measurements.

A hybrid deep model with HOG features for Bangla handwritten numeral classification

Considering the practical significances, handwriting recognition is getting an intense interest to the research community. Through, several studies have been conducted for Bengali handwriting recognition, a robust model for Bengali numerals classification is still due. Therefore, a hybrid model is presented in this paper, which aims to classify the Bengali numerals more precisely. The proposed model bridges hand crafted feature extraction based approaches with the automatically learnt features of Convolutional Neural networks (CNN). It is observed that the proposed model outperforms existing models with lesser epochs. The proposed model is trained and tested with the ISI numeral dataset and also cross-validated with the CAMTERDB numeral dataset. For both scenarios, proposed model shows consistency and demonstrate the maximum accuracy of 99.02% and 99.17%, respectively. For the CMATERDB collection, the proposed model achieves the best accuracy rate reported till date.

Learning “initial feature weights” for CBIR using query augmentation

Content-based image retrieval (CBIR) is one of the most active fields of research in image processing and information retrieval. In CBIR, an image is given as query, instead of text, and a set of relevant images is returned as an output. Researchers have generally used multiple image features in conjunction to achieve high CBIR performance. Relevance feedback has also seen widespread adoption as technique to utilise user feedback to further refine search results. In this paper, we propose a technique to ascertain the initial feature weights before the actual query is processed. The weights of the features are determined by augmenting the query image through different transformations of the query image. The proposed method is tested on the VisTex and Outex_TR_00000 texture collections. The performance is measured by average retrieval rate, precision and recall. Our results do not show any degradation on retrieval performance on collections that have relevance classes which are generally uniform. However, on collections that are more heterogeneous, our proposed method leads to better search results.