D. R. Ramesh Babu

Recognition of machine printed broken characters based on gradient patterns and its spatial relationship

Recognition of degraded characters is a challenging problem in the field of document image analysis. Two main reasons for degradation of characters are due to noise scanning and intrinsic degradation caused by font variations. The degradation of characters is mostly in the form of characters being broken at several places which hinders their recognition of OCR systems. Many OCRs have been designed which correctly identify fine printed characters without any anomalies. However, very few research works has been reported on the recognition of the degraded character recognition due to its complexity. The efficiency of the OCRs system decreases if the input image is degraded and the characters are broken at several places, which frequently occur in old documents. In this paper, a novel approach based on slope pattern and its spatial relationship is used for recognizing broken characters. The proposed method is based on the computation of slope pattern of the individual characters in eight view-directions and generating image code for each character. The algorithm has been applied on the broken characters with successful results.

Application of Region of Interest Compressed Sensing to accelerate magnetic resonance angiography.

Magnetic Resonance Angiography (MRA) is a group of techniques based on Magnetic Resonance Imaging (MRI) to image blood vessels. Compressed Sensing (CS) is a mathematical framework to reconstruct MR images from sparse data to minimize the data acquisition time. Image sparsity is the key in CS to reconstruct MR images. CS technique allows reconstruction from significantly fewer k-space samples as compared to full k-space acquisition, which results in reduced MRI data acquisition time. The images resulting from MRA are sparse in native representation, hence yielding themselves well to CS. Recently our group has proposed a novel CS method called Region of Interest Compressed Sensing (ROICS) as a part of Region of Interest (ROI) weighted CS. This work aims at the implementation of ROICS for the first time on MRA data to reduce MR data acquisition time. It has been demonstrated qualitatively and quantitatively that ROICS outperforms CS at higher acceleration factors. ROICS technique has been applied to 3D angiograms of the brain data acquired at 1.5T. It helps to reduce the MRA data acquisition time and improves the visualization of arteries. ROICS technique has been applied on 4 brain angiogram data sets at different acceleration factors from 2× to 10×. Reconstructed images show ROICS technique performs better than conventional CS technique and is quantified by the comparative Signal to Noise Ratio (SNR) in the ROI.

Automatic detection and tracking of moving objects in complex environments for video surveillance applications

Autonomous video surveillance and monitoring has a rich history. A new method for detecting multiple moving objects based on improved background subtraction model and for tracking is based on feature based approach has proposed. Then identified moving objects are also counted, by indexing individually. The proposed algorithm is automatic and efficient in intelligent surveillance applications like vehicles monitoring, event recognition, and crime prevention, etc. The proposed model has proved to be robust in various environments (including indoor and outdoor scenes) and different types of background scenes. Experiments on real scenes show that the algorithm is effective for object detection and tracking.

LoG-DEM: Log Gabor filter and Dominant Eigen Map approaches for moving object detection and

In recent years, the numbers of Visual Surveillance systems have greatly increased, and these systems have developed into intellectual systems that automatically detect, track, and recognize objects in video. Automatic moving object detection and tracking is a very challenging task in video surveillance applications. In this regard, many methods have been proposed for Moving Object Detection and Tracking based on edge, color, texture information. Due to unpredictable characteristics of objects in foggy videos, the task of object detection remains a challenging problem. In this paper, we propose a novel scheme for moving object detection based on Log Gabor filter (LGF) and Dominant Eigen Map (DEM) approaches. Location of the moving object is obtained by performing connected component analysis. In turn, a Moving Object is Tracked based on the centroid manipulation. Number of experiments is performed using indoor and outdoor video sequences. The proposed method is tested on standard PETS datasets and many real time video sequences. Results obtained are satisfactory and are compared with existing well known traditional methods.

Scale weight selection for feature extraction using complex wavelets: A framework

Unmanned Air Vehicles (UAVs) have become an intelligent asset for surveillance, target tracking and reconnaissance in both urban and battlefield settings. This paper gives a framework for scale weight selection during feature extraction in aerial images from UAV. Dual-Tree Complex Waveform technique is used to extract rich feature descriptors of keypoints in images so that full phase and amplitude information can be retained at each scale. The scale weights are dependent on image characteristics such as the illumination and the contrast levels. The outcome of the framework shows promising results in terms of less redundancy of salient features from the images and hence improving the computational speed.

Rotation-Invariant Fast Feature Based Image Registration for Motion Compensation in Aerial Image Sequences

Motion compensation can be used as a preprocessing step in the application of object tracking in aerial image sequences from Unmanned Air Vehicle to cancel the effect of camera motion. In this paper, we demonstrate Aerial Image Registration that gives high degree of accuracy for motion compensation. Rotation Invariant Fast Features that use approximate radial gradient transform are used to reduce the computation time of feature extraction considerably. These descriptors well define the aerial image features taken from platforms like UAV that are prone to high degree of rotation due to sudden maneuver, scaling, illumination change and noise. Another contribution of the paper is in the formulation of new framework for set based registration of aerial images. Results using the group scheme outperform the usual pair wise registration and demonstrate real-time performance.

A survey on noise reduction techniques for lung cancer detection

Lung cancer is one of the fatal diseases that mainly affect the pulmonary nodules of the lungs. Examination of image is presently an essential step of the lung diseases processes like diagnostic, prognostic and follow-up. The survival rate of lung cancer is very low when compared with all other types of cancer. The need for identifying lung cancer at an early stage is very essential and is an active research area in the field of medical image processing. Several Computer aided systems have been intended to distinguish the lung cancer at its initial stage. Various types of images are used for detection of lung diagnosis. The most important challenging task is detection of lung nodule. Computed Tomography (CT) images are generally chosen due to less distortion, low noise, better clarity, less time consumption and low cost. There are different types of the noise present in the images we obtain for the lung mass detection like impulse noise, Gaussian noise and speckle noise. Removal of noise from images is the most active field of research. This paper presents the review on the lung cancer, types of noise in medical imaging and then the methods for the removal of noise.

Automatic Classification of Lung Nodules into Benign or Malignant Using SVM Classifier

Carcinoma of lungs is allied to the cancers that are causing the highest number of deaths all over the world. It is very important to improvise the detection methods so that the rate of survival can be increased. In this paper, new algorithm has been proposed to segment the lung regions using Active Contour method. Once the detection of nodules is through and Gray level Co-occurrence Matrix (GLCM) is used to calculate the texture features. HARALICK texture features are calculated and dominant features are extracted. Support Vector Machine (SVM) Classification of the nodules is done using SVM classifier. Satisfactory results have been obtained. Lung CT scan images are taken from LIDC-IDRI database.

Automatic Extraction of Lung Regions Using Curve Analysis for the Detection of Lung Cancer in CT Scans

Segmentation of lung regions with lung nodules at mediastinum is the first step in computer-aided detection (CAD) which provides a better diagnosis of lung cancer. The existing methods fail in segmentation of lung regions with the cancer tumors at the mediastinum of the lungs. In this paper, a new approach is proposed that extracts lung regions with cancer tumors at the mediastinum of the lungs based on curve analysis. The proposed algorithm is tested on 05 patient’s dataset which consists of 60 images of the Lung Image Database Consortium (LIDC) and the results are found to be satisfactory with 99 % average overlap measure (AΩ). The proposed algorithm extracts lung nodules at the mediastinum of the lungs which helps in detection of lung cancer.

Analysis of Estimation Techniques for Feature Based Navigation of Unmanned Air Vehicles

This paper investigates different linear and non-linear state estimation techniques applicable in the scenario of unmanned air vehicle platform with an image sensor that attempts to navigate in a known urban terrain. Feature detection and matching steps are carried out using DTCWT (Dual tree Complex Wavelet Transform) descriptor. The state parameters of the vehicle are subsequently estimated using the measurements from the image sensor. Due to the non-linear nature of the problem, non-linear filters such as particle filters are often used. Various other methods have been proposed in the literature to the problem ranging from linear Kalman filter to non-linear probabilistic based techniques. This paper therefore attempts to study different state estimation methods and understand two main approaches in particular the Kalman and particle filter for the problem.