Key Papers


A sampling approach to ultra-fast object location
Davies, E.R.
Real-Time Imaging, 7, no. 4, 339–355 (2001)

This paper presents new work to determine lowest bounds on the computation needed to locate objects in digital images. While it was originally applied to the rapid location of cereal grains for real-time inspection, the concepts are now being found useful as a rigorous starting point for understanding general visual search, and have an interesting relationship with the saccades of the human visual system. During 2007 this generalisation has been published in an Electronics Letter (43(9), 508), and slightly more fully in VIE 2007.


Design of efficient line segment detectors for cereal grain inspection
Davies, E.R., Bateman, M., Mason, D.R., Chambers, J. and Ridgway, C.
Pattern Recogn. Lett., 24, nos. 1–3, 421–436 (2003)

This paper arose from a grant from the Home-Grown Cereals Authority to carry out research in conjunction with DEFRA. It presents the design of sensitive, robust, real-time algorithms for locating insects amongst cereal grains. The algorithms take the form of efficient bar detectors, and have been shown to be more effective than detectors based on edge detection. They generalise to use as streak detectors, and significant interest has been shown in their use for totally different applications, such as locating veins in human retinal images. Thus the methodology is highly generic, applying well outside the area of original motivation.


An analysis of the geometric distortions produced by median and related image processing filters
Davies, E.R.
invited review in Advances in Imaging and Electron Physics, 126, 93–193 (2003)

This paper encapsulates and significantly extends Roy's work in understanding the exact extent of the geometric distortions produced by median and related filters (whose aim is largely to eliminate noise from digital images). While a continuum approximation gave useful results, later discrete structural analysis eventually led to essentially exact agreement between theory and experiment, so that the subject could be regarded as satisfactorily closed, at least for grey-scale images. This work includes theoretical models and experimental tests of the shifts produced by the whole family of rank-order filters, and as such has strong relevance and importance for mathematical morphology.


Mode filters and their effectiveness for processing colour images
Charles, D. and Davies, E.R.
Imaging Science, 52, no. 1, 3–25 (2004)

This paper investigates the properties of mode filters in grey and colour images, and extends the work of the previous paper, (a) to the colour domain and (b) to provide detailed explanations of the performance of mode filters. Somewhat surprisingly, it also shows astonishing performance when applied to colour images with up to 70% impulse noise, removing the noise almost completely and amounting to a breakthrough for the low computation colour mode filter that had been devised in this work. Hitherto, mode filters seemed to be mainly useful for image enhancement, but their use for gross noise suppression is entirely new.


Stable bi-level and multi-level thresholding of images using a new global transformation
Davies, E.R.
IET Computer Vision, 2, no. 2, Special Issue on Visual Information Engineering, Valestin, S. (ed.), 60–74 (2008)

This paper describes a new transformation for finding global valleys in 1D distributions, and is particularly useful for thresholding grey-scale images. It estimates the global significance of all the valleys that are located, and can therefore cope with multi-mode distributions. One of the main advantages of the resulting global valley method (GVM) is that it permits partially hidden minima to be reliably located without complicated analysis. Overall, the GVM is demonstrated to have high sensitivity for the detection of subsidiary minima – including those arising from practically important image detail (such as defects or contaminants in an automated inspection scenario).


The application of machine vision to food and agriculture: a review
Davies, E.R.
Imaging Science, 57, no. 4, 197–217 (2009)

This paper reviews developments since 2000 in the application of machine vision to food and agriculture. While visible light frequently provides enough useful information to make sound judgements, X-rays and hyperspectral imaging have been increasingly valuable. There have also been valuable developments in the use of 3D methods, such as ‘double Hough transforms’ for the accurate delineation of crop rows, so that ‘precision agriculture’ can be realized, and the use of sets of visual calibration points so that robot vehicles can determine their exact locations and headings. Overall, the steady development of useful vision algorithms has been well matched by the capability of today’s computers to implement them at sufficiently high speeds to make them viable.


Improved line detection algorithm for locating road lane markings
Mastorakis, G. and Davies, E.R.
Electronics Letters, 47, no. 3, 183–184 (2011)

This paper describes an improved line detection algorithm for locating road lane markings. After scanning images to find mid-points of potential lane markings, they are passed to the line detection algorithm, which selects the line with the greatest support. After finding each line, the algorithm deletes not only the inliers within a ‘fit tolerance’ distance df but also the innermost outliers lying within a larger ‘delete tolerance’ distance dd. This procedure is found to increase robustness because elimination of remanent points near each line prevents them from confusing the algorithm when it searches for further lines.


Object location using the Hough transform
Davies, E.R.
Chapter 18 in Handbook of Machine Vision, Batchelor, B.G. and Miller, J.W.V. (eds.), Springer Verlag, London,
773–800 (2012)

This chapter finds that inference methods often provide a significantly more robust means for locating complex objects, the Hough transform being one of the most important examples of such methods. Graph matching also has this property and demonstrates certain equivalences to the Hough transform technique. In fact, spatial matched filtering is intrinsically the most sensitive method for locating objects, and this is shown to explain the success of both the Hough transform and graph matching. Nevertheless, careful algorithm design is needed to reduce the underlying computational difficulties of this approach.


Computer vision for automatic sorting in the food industry
Davies, E.R.
Chapter 6 in Computer Vision Technology in the Food and Beverage Industries, Sun, D.-W. (ed.), Woodhead Publishing Ltd.,
Cambridge, UK, 150–180 (2012)

This chapter describes how automatic sorting has been carried out increasingly widely in the food industry via the application of computer vision. This has been ensured by continual developments in image processing and pattern recognition. At the same time, advances in computer technology have permitted viable implementations to be achieved at lower cost. However, progress is being held up by the need for tailored development, so future algorithms will have to be made much more trainable than at present. In addition, recent methodologies such as hyperspectral imaging should become accepted in more application areas – particularly the assessment of raw fruit and vegetable produce.


The dramatically changing face of computer vision
Davies, E.R.
Chapter 1 in Advanced Methods and Deep Learning in Computer Vision, E.R. Davies and Matthew Turk (eds.), Elsevier, Oxford, UK, 1–91 (2022)

This chapter aims to explain the concepts leading up to the recently evolved deep learning milieu, covering aspects such as image processing, feature detection, object recognition, segmentation and tracking: by providing a useful level of background theory, and an introduction to deep learning, the chapter aims to help prepare readers for the advanced chapters that are to follow.