Andrew Busch

Texture for script identification

The problem of determining the script and language of a document image has a number of important applications in the field of document analysis, such as indexing and sorting of large collections of such images, or as a precursor to optical character recognition (OCR). In this paper, we investigate the use of texture as a tool for determining the script of a document image, based on the observation that text has a distinct visual texture. An experimental evaluation of a number of commonly used texture features is conducted on a newly created script database, providing a qualitative measure of which features are most appropriate for this task. Strategies for improving classification results in situations with limited training data and multiple font types are also proposed.

Logarithmic quantisation of wavelet coefficients for improved texture classification performance

The coefficients of the wavelet transform have been widely used for texture analysis tasks, including segmentation, classification and synthesis. Second order statistics of such values have been shown to give excellent performance in these applications, and are typically calculated using co-occurrence matrices, which require quantisation of the coefficients. In this paper, we propose a non-linear quantisation function which is experimentally shown to better characterise textured images, and use this to formulate a new set of texture features, the wavelet log co-occurrence signatures.

Texture classification using wavelet scale relationships

It has been documented in the literature that texture can be well characterised by features obtained from its multi-scale representation. Typically, the textured image being analysed is decomposed into separate frequency and/or orientation bands, and features extracted separately from each such band. In this paper, we propose that features modelling the relationships between scale bands of such a representation provide a better characterisation of textured images than features extracted from individual bands alone. Using this conjecture, we develop a novel feature set for texture classification, and demonstrate its effectiveness using a set of images obtained from the Brodatz texture album.

Multi-font script identification using texture-based features

The problem of determining the script and language of a document image has a number of important applications in the field of document analysis, such as indexing and sorting of large collections of such images, or as a precursor to optical character recognition (OCR). In this paper, we investigate the use of texture as a tool for determining the script of a document image, based on the observation that text has a distinct visual texture. An experimental evaluation of a number of commonly used texture features is conducted on a newly created script database, providing a qualitative measure of which features are most appropriate for this task. Strategies for improving classification results in situations with limited training data and multiple font types are also proposed.

An image processing approach for estimating the number of live prawn larvae in water

We address the problem of accurate estimation of prawn larvae numbers in small water-filled containers using image processing techniques. Images of the containers are captured from sampled video signals of a camera suspended over the containers. The images are preprocessed for noise removal and enhancement. The resulting images are then processed to estimate the number of prawn larvae with an acceptable accuracy. Our preliminary results show that estimates of prawn larvae numbers with accuracy of between 90-95% are achievable under controlled conditions. Some issues of system development and practical implementation are discussed.

A Method to Create Stable Lighting and Remove Specular Reflections for Vision Systems

A lighting system and method has been developed which has shown in testing to allow quality images to be obtained that are free from two particular problems, specular reflections on the subject, and light intensity variation. These problems both diminish the ability to compare objects for attributes such as colour variation, edges, contours, and many other features. The system developed eliminates specular reflection by using the cross-polarisation configuration, and reduced flickering due to fluctuations in the power supply to negligible levels by constructing a high-power DC source capable of providing sufficient 12 Volt power. These two improvements create an environment suitable for taking high-quality, noise free images at high shutter speeds for the purpose of assessing the quality of strawberries moving on a real-time production line.

Segmentation of the Left Ventricle in Echocardiography Using Contextual Shape Model

Accurate localization of the left ventricle (LV) boundary from echocardiogram images is of vital importance for the diagnosis and treatment of heart disease. Statistical shape models such as active shape models (ASM) have been commonly used to perform automatic detection of this boundary. Such models perform well when there is low variability in the underlying shape subspace and an accurate initialization can be provided, however in the absence of these conditions results are often much poorer. In the case of LV echocardiogram images, such variability is often encountered in patients with abnormal LV function. In this paper we propose a fully automatic segmentation technique using deep learning in a Bayesian nonparametric framework. Our model uses a dynamic statistical shape model comprised of training shapes from select weighted subsets of the feature subspace. Subsets are chosen during the iterative segmentation process according to a latent temporal component allocation variable, determined from joint deep features and LV landmark information using a Dirichlet process mixture model with Chinese restaurant process prior. Testing is performed on a data set comprising images of the LV acquired from patients exhibiting both normal and abnormal LV function, and the results using our technique compared to both the ASM and other state of the art techniques. Results from this testing show an improvement in the LV localization accuracy, particularly when LV function is abnormal.

Segmentation of the Left Ventricle from Ultrasound Using Random Forest with Active Shape Model

This paper presents a model-based learning segmentation algorithm to detect the left ventricle (LV) boundary of the heart from ultrasound (US) images by combining a random forest classifier with an active shape model (ASM). Our method applies an ASM for initial detection of the LV landmarks. Each landmark is subsequently directed radially inward or outward as a result of the random forest classifier identifying the landmark as outside or inside the LV boundary, respectively. This is done while preserving the shape characteristics obtained from the ASM. Our objective is to evaluate the combined application of a random forest classifier with an ASM for detecting the LV boundary with US images. Accuracy of this method is evaluated by comparing both our method and ASM to LV contours traced by an expert. A dataset of 85 randomly selected patient studies was chosen. The method exhibits improved accuracy compared to the ASM, producing a global overlap coefficient of 90.09% compared to 83.8% obtained with an active shape model.

An image processing approach to distance estimation for automated strawberry harvesting

In order to successfuly navigate between rows of plants, automated strawberry harvesters require a robust and accurate method of measuring the distance between the harvester and the strawberry bed. A diffracted red laser is used to project a straight horizontal line onto the bed, and is viewed by a video camera positioned at an angle to the laser. Using filtering techniques and the Hough transform, the distance to the bed can be calculated accurately at many points simultaneously, allowing the harversters navigation system to determine both its position and angle relative to the bed. Testing has shown that this low-cost solution provides near-perfect field performance.