Showing items 1 - 5 of 5

Your selections:

Cuboid colour image segmentation using intuitive distance measure

- Tania, Sheikh, Murshed, Manzur, Teng, Shyh, Karmakar, Gour

Hierarchical colour image segmentation by leveraging RGB channels independently

- Tania, Sheikh, Murshed, Manzur, Teng, Shyh, Karmakar, Gour

  • Authors: Tania, Sheikh , Murshed, Manzur , Teng, Shyh , Karmakar, Gour
  • Date: 2019
  • Type: Text , Conference paper
  • Relation: 9th Pacific-Rim Symposium on Image and Video Technology, PSIVT 2019 Vol. 11854 LNCS, p. 197-210
  • Full Text:
  • Reviewed:
  • Description: In this paper, we introduce a hierarchical colour image segmentation based on cuboid partitioning using simple statistical features of the pixel intensities in the RGB channels. Estimating the difference between any two colours is a challenging task. As most of the colour models are not perceptually uniform, investigation of an alternative strategy is highly demanding. To address this issue, for our proposed technique, we present a new concept for colour distance measure based on the inconsistency of pixel intensities of an image which is more compliant to human perception. Constructing a reliable set of superpixels from an image is fundamental for further merging. As cuboid partitioning is a superior candidate to produce superpixels, we use the agglomerative merging to yield the final segmentation results exploiting the outcome of our proposed cuboid partitioning. The proposed cuboid segmentation based algorithm significantly outperforms not only the quadtree-based segmentation but also existing state-of-the-art segmentation algorithms in terms of quality of segmentation for the benchmark datasets used in image segmentation. © 2019, Springer Nature Switzerland AG.

An Enhanced Local Texture Descriptor for Image Segmentation

- Tania, Sheikh, Murshed, Manzur, Teng, Shyh, Karmakar, Gour

  • Authors: Tania, Sheikh , Murshed, Manzur , Teng, Shyh , Karmakar, Gour
  • Date: 2020
  • Type: Text , Conference paper
  • Relation: 2020 IEEE International Conference on Image Processing, ICIP 2020 Vol. 2020-October, p. 1526-1530
  • Full Text: false
  • Reviewed:
  • Description: Texture is an indispensable property to develop many vision based autonomous applications. Compared to colour, feature dimension in a local texture descriptor is quite large as dense texture features need to represent the distribution of pixel intensities in the neighbourhood of each pixel. Large dimensional features require additional time for further processing that often restrict real-time applications. In this paper, a robust local texture descriptor is enhanced by reducing feature dimension by three folds without compromising the accuracy in region-based image segmentation applications. Reduction in feature dimension is achieved by exploiting the mean of neighbourhood pixel intensities radially along lines across a certain radius, which eliminates the need for sampling intensity distribution at three scales. Both the results of benchmark metrics and computational time are promising when the enhanced texture feature is used in a region-based hierarchical segmentation algorithm, a recent state-of-the-art technique. © 2020 IEEE.

Efficient texture descriptors for image segmentation

- Tania, Sheikh

  • Authors: Tania, Sheikh
  • Date: 2022
  • Type: Text , Thesis , PhD
  • Full Text:
  • Description: Colour and texture are the most common features used in image processing and computer vision applications. Unlike colour, a local texture descriptor needs to express the unique variation pattern in the intensity differences of pixels in the neighbourhood of the pixel-of-interest (POI) so that it can sufficiently discriminate different textures. Since the descriptor needs spatial manipulation of all pixels in the neighbourhood of the POI, approximation of texture impacts not only the computational cost but also the performance of the applications. In this thesis, we aim to develop novel texture descriptors, especially for hierarchical image segmentation techniques that have recently gained popularity for their wide range of applications in medical imaging, video surveillance, autonomous navigation, and computer vision in general. To pursue the aim, we focus in reducing the length of the texture feature and directly modelling the distribution of intensity-variation in the parametric space of a probability density function (pdf). In the first contributory chapter, we enhance the state-of-the-art Weber local descriptor (WLD) by considering the mean value of neighbouring pixel intensities along radial directions instead of sampling pixels at three scales. Consequently, the proposed descriptor, named Radial Mean WLD (RM-WLD), is three-fold shorter than WLD and it performs slightly better than WLD in hierarchical image segmentation. The statistical distributions of pixel intensities in different image regions are diverse by nature. In the second contributory chapter, we propose a novel texture feature, called ‘joint scale,’ by directly modelling the probability distribution of intensity differences. The Weibull distribution, one of the extreme value distributions, is selected for this purpose as it can represent a wide range of probability distributions with a couple of parameters. In addition, gradient orientation feature is calculated from all pixels in the neighbourhood with an extended Sobel operator, instead of using only the vertical and horizontal neighbours as considered in WLD. The length of the texture descriptor combining joint scale and gradiet orientation features remains the same as RM-WLD, but it exhibits significantly improved discrimination capability for better image segmentation. Initial regions in hierarchical segmentation play an important role in approximating texture features. Traditional arbitrary-shaped initial regions maintain the uniform colour property and thus may not retain the texture pattern of the segment they belong to. In the final contributory chapter, we introduce regular-shaped initial regions by enhancing the cuboidal partitioning technique, which has recently gained popularity in image/video coding research. Since the regions (cuboids) of cuboidal partitioning are of rectangular shape, they do not follow the colour-based boundary adherence of traditional initial regions. Consequently, the cuboids retain sufficient texture pattern cues to provide better texture approximation and discriminating capability. We have used benchmark segmentation datasets and metrics to evaluate the proposed texture descriptors. Experimental results on benchmark metrics and computational time are promising when the proposed texture features are used in the state-of-the-art iterative contraction and merging (ICM) image segmentation technique.
  • Description: Doctor of Philosophy

A robust local texture descriptor in the parametric space of the weibull distribution

- Tania, Sheikh, Karmakar, Gour, Teng, Shyh, Murshed, Manzur

  • Authors: Tania, Sheikh , Karmakar, Gour , Teng, Shyh , Murshed, Manzur
  • Date: 2023
  • Type: Text , Journal article
  • Relation: IEEE Transactions on Multimedia Vol. 25, no. (2023), p. 6053-6066
  • Full Text: false
  • Reviewed:
  • Description: Research in texture feature approximation is still in the embryonic stage because of difficulties in developing a sound theoretical model to express the unique pattern in the intensity-variation of pixels in the neighbourhood of the pixel-of-interest so that it can sufficiently discriminate different textures. Local texture descriptors are widely used in image segmentation as they comprise pixel-wise features. The Weber local descriptor (WLD) with differential excitation and gradient orientation components, inspired by Weber's Law, has been leveraged in the state-of-the-art iterative contraction and merging (ICM) image segmentation technique. However, WLD has inherent drawbacks in the formulation of the components that limit its discriminatory capability. This paper introduces a novel texture descriptor by directly modelling the distribution of intensity-variation in the parametric space of the Weibull distribution using its shape and scale parameters. A unified 'joint scale' texture property is introduced, which can discriminate textures better than the individual parameters while keeping the length of the descriptor shorter. Additionally, the accuracy of WLD's gradient orientation component is improved by using an extended Sobel operator and expressing gradients in -
  • 1