#### DMCA

## Unsupervised Recognition of Salient Colour for Real-Time Image Processing

Citations: | 5 - 2 self |

### Citations

4454 |
Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations
- Witten, Frank
- 1999
(Show Context)
Citation Context ...(i.e. Y CbCr pixel coordinates, also scaled between −1 and 1 for each axis) belonging to that respective class. Concretely, a MATLAB implementation of the LIBSVM one-class SVM implementation was used =-=[11]-=-. Despite the fact that previous research has indicated the implementation of such a generalisation process improves LUT performance [9], a course-grain grid search of SVM parameters yielded consisten... |

2391 | Mean shift: A robust approach toward feature space analysis
- COMANICIU, MEER, et al.
- 2002
(Show Context)
Citation Context ...arameterises the underlying distribution as a superposition of hyperspherical distributions; expectation maximisation [1], which generalises k-means by assuming a mixture of Gaussians; and mean shift =-=[5]-=-, which generates a non-parametric model of the entire distribution by convolving all feature vectors by some kernel function. 3.1 k-Means Clustering Given a set of m data points (corresponding with f... |

1062 |
Pattern recognition and machine learning
- Bishop
- 2006
(Show Context)
Citation Context ...0]. Specifically, three algorithms are considered: k-means clustering [6], which parameterises the underlying distribution as a superposition of hyperspherical distributions; expectation maximisation =-=[1]-=-, which generalises k-means by assuming a mixture of Gaussians; and mean shift [5], which generates a non-parametric model of the entire distribution by convolving all feature vectors by some kernel f... |

548 |
On the generalized distance in statistics
- Mahalanobis
- 1936
(Show Context)
Citation Context ...d to k-means, which associated each data point to a respective label via the nearest neighbours method (i.e. minimum Euclidean distance), expectation maximisation (EM) utilises a Mahalanobis distance =-=[8, 10]-=- d(x(i), µk;Σk) = (x (i) − µk) T Σ −1 k (x (i) − µk), where x(i) and µk are data points and cluster centroids respectively (x (i), µk ∈ R n), and Σk are their corresponding covariance estimates. As op... |

303 | Robocup: The robot world cup initiative
- Kitano, Asada, et al.
- 1997
(Show Context)
Citation Context ...onstant time access. This work focuses on this reduced task of colour segmentation, within an environment where salient features exhibit some significant degree of colour-coding. RoboCup robot soccer =-=[7]-=- is chosen as such as scenario for experimentation, where field lines, goals and the ball are each assigned unique colours, and where maintaining real-time processing performance is critical for robot... |

299 |
Algorithm as 136: A k-means clustering algorithm
- Hartigan, Wong
- 1979
(Show Context)
Citation Context ...sampled from some unknown underlying probability distribution, and attempts to locate clusters (modes) within this distribution [10]. Specifically, three algorithms are considered: k-means clustering =-=[6]-=-, which parameterises the underlying distribution as a superposition of hyperspherical distributions; expectation maximisation [1], which generalises k-means by assuming a mixture of Gaussians; and me... |

252 | Computer Vision: Algorithms and Applications
- Szeliski
- 2010
(Show Context)
Citation Context ...nual colour mapping. Keywords: Computer vision, colour vision, robotics, RoboCup 1 Introduction Szeliski describes image segmentation simply as the task of finding groups of pixels that “go together” =-=[10]-=-. This is an abstract notion that corresponds with an inherently subconscious human process: the ability to look at an image and identify salient features, such as a person, landmark or household item... |

197 | Color image segmentation : advances and prospects
- Cheng, Jiang, et al.
(Show Context)
Citation Context ... and external validation techniques [2]. Similarly to HSV, Y CbCr separates chrominance information into two channels Cb (blue chroma) and Cr (red chroma), and intensity into a third channel Y (luma) =-=[4, 10]-=-. The Y CbCr colour space can be obtained applying the following linear transformation to the RGB space Y ′Cb Cr = 0.299 0.587 0.114−0.168736 − 0.331264 0.5 0.5 −0.418688 −0.081312 R... |

15 | Application of SVMs for colour classification and collision detection with AIBO robots.
- Quinlan, Chalup, et al.
- 2003
(Show Context)
Citation Context ...ementation of the LIBSVM one-class SVM implementation was used [11]. Despite the fact that previous research has indicated the implementation of such a generalisation process improves LUT performance =-=[9]-=-, a course-grain grid search of SVM parameters yielded consistent reduction in performance. This discrepancy may be influenced by a number of factors: the improved quality of input LUTs, the smoother ... |

9 | Evaluation of colour models for computer vision using cluster validation techniques
- Budden, Fenn, et al.
- 2013
(Show Context)
Citation Context ... Past research has demonstrated the Y CbCr colour space as optimal (among the examples listed above) for unsupervised colour segmentation, in terms of both internal and external validation techniques =-=[2]-=-. Similarly to HSV, Y CbCr separates chrominance information into two channels Cb (blue chroma) and Cr (red chroma), and intensity into a third channel Y (luma) [4, 10]. The Y CbCr colour space can be... |

8 | A novel approach to ball detection for humanoid robot soccer
- Budden, Fenn, et al.
- 2012
(Show Context)
Citation Context ...presented in Y CbCr format. 2.2 Colour Look-Up Tables In computer vision, a mapping from an arbitrary 3-component colour space C to a set of coloursM assigns a class label mi ∈M to every point cj ∈ C =-=[3]-=-. If each channel is represented by an n-bit value and k = |M | represents the number of defined class labels, then C →M, where C = {0, 1, . . . , 2n − 1}3 and M = {m0,m1, . . . ,mk−1} . Concretely, i... |