Adaptive Hough transform for the detection of natural shapes under weak affine transformations

Abstract

This paper introduces a two-steps adaptive generalized Hough transform (GHT) for the detection of non-analytic objects undergoing weak affine transformations in images. The first step of our algorithm coarsely locates the region of interest with a GHT for similitudes. The returned detection is then used by an adaptive GHT for affine transformations. The adaptive strategy makes the computation more amenable and ensures high accuracy, while keeping the size of the accumulator array small. To account for the deformable nature of natural objects, local shape variability is incorporated into the algorithm in both the detection and reconstruction steps. Finally, experiments are performed on real medical data showing that both accuracy and reasonable computation times can be reached.

Deformation tolerant generalized Hough transform for sketch-based image retrieval in complex scenes

“…Concerning GHT generalizations, Wolfson [3] proposes to extend the original GHT to deal with articulated objects. In [15] the authors use a set of shape examples to train the system to recognize a class of objects. The R-Table is enriched with information describing the statistical distribution of the local shape variability of the class prototype silhouette and such information is used at voting time to compute the amount of smoothing according to the data.…”

“…As mentioned in Section 3, vote spreading has been largely used by different authors to deal with data uncertainty and quantization problems [2,5] or to extend the GHT to treat classification issues [6,15]. In [6] the authors propose to smooth the vote using an exponential-based distance function.…”

“…In [6] the authors propose to smooth the vote using an exponential-based distance function. In [15] the local shape variability of an object class is represented by associating each entry of the RTable with a statistical distribution, used at voting time to adapt the amount of smoothing. In [5] the shape of the spreading area and the bin increment value depend on the pair of features (image, model) selected for each single vote action and on a fuzzy representation of the estimated data uncertainty.…”

Estimation of muscle pennation angle in ultrasound images using the beamlet transform

“…However, the classical edge detectors (such as Canny, Sobel and Prewitt operators) seem to be inappropriate in ultrasound images due to the speckle noise. Hough transform is one of the most powerful approaches for the detection and recognition of patterns with known simple geometrical shapes in images, especially those with the line and circle shapes [9] . It has been used to estimate the fiber orientation with large computational costs [7] .…”

Automatic detection of supporting device positioning in intensive care unit radiography

“…In this study, a method of automatic detection for tubes such as an endotracheal tube (ET) or a food tube (FT) was researched in order to improve the quality of patient management in the intensive care unit (ICU). For linear features detection, the Hough transform is commonly used (8,9). It is one of the most fundamental tasks in pattern recognition (10).…”