Image stitching, the technique of merging multiple images with overlapping regions, has various applications in real-life scenarios such as security systems, autonomous vehicles, and sports analytics. Several image stitching methodologies have been developed, resulting in a wide range of traditional feature-based and deep learning approaches. In this paper, we utilise the Adobe Panoramic Dataset to assess eight approaches for stitching four overlapping images. We develop six novel feature-based stitching approaches, including AKAZE-SIFT-BF KNN-RANSAC. As an example approach, KAZE-SIFT-BF KNN-RANSAC utilises Accelerated KAZE (AKAZE) for feature detection, Scale Invariant Feature Transformation (SIFT) to generate feature descriptors, Brute Force with k-Nearest Neighbours and a ratio test for keypoint matching, and Random Sample Consensus (RANSAC) for matched keypoint outlier removal. Evaluation is performed using the CombSIQE metric and qualitative comparison to evaluate the quality of the resulting stitched images. Performance evaluation shows that our proposed approaches produce comparable quality stitched images to the optimised SOTA SuperPoint and SuperGlue approach in outdoor scenes, while remaining consistent in indoor scenes.

Scale-invariant interest point detection is crucial for many computer vision tasks in dynamic environments, such as manufacturing, where identifying recurring visual landmarks is essential for process monitoring. However, existing solutions, including the Scale-Invariant Feature Transform and its descendants, incur high computational costs due to operations across multiple scales. Drawing inspiration from the Finite Element Scale-Invariant Detector, this paper introduces a novel square-spiral derivative that integrates a vectorized image addressing scheme with complementary processing techniques to enhance runtime performance. Experimental evaluations demonstrate that the proposed approach remains robust under moderate geometric and photometric transformations, achieving competitive performance compared to widely used detectors in conditions that resemble those in controlled dynamic environments. Furthermore, it significantly reduces computational overhead, providing an efficient solution for vision-based manufacturing tasks.

This paper presents an efficient approach to corner detection for images using a spiral addressing scheme in conjunction with simulated, biological involuntary eye movements. As part of this approach, a combined gradient detection and smoothing operation is used to quickly obtain a feature representation that can be used with a standard 'cornerness' measure. A computationally efficient use of a spiral address scheme to apply further processing operations such as non-maximum suppression is demonstrated. An evaluation of three corner detection methods is presented and results demonstrate that a method designed for a spiral based, biologically inspired approach can achieve a significantly faster runtime than comparative methods designed for a traditional approach.

Efficient processing of digital images is a key consideration in many machine vision tasks. Traditional image processing approaches often struggle to meet this demand, particularly at the initial low-level of processing image pixels. To overcome this, we propose a spiral based processing approach which takes inspiration from the asymmetric lattice of interlocking cells found in the human visual system. Here we demonstrate the efficiency of the proposed spiral approach for multi-scale feature extraction. This is complemented by a biologically inspired image acquisition process which is used to capture nine image frames at different spatial locations. The results demonstrate that the biologically inspired spiral approach offers a faster alternative to corresponding traditional image processing approaches.

Fast image processing is a key element in achieving real-time image and video analysis. Here, a novel framework based on a spiral architecture is used to facilitate fast image processing, in particular, fast corner detection. Unlike a conventional image addressing scheme where the picture elements are indexed using two-dimensional Cartesian coordinates, a spiral addressing scheme enables the image to be stored and indexed as a one-dimensional vector. Image processing is hastened through the combined use of the one-dimensional structure and a lookup table. The performance is evaluated by the application of a corner detector based on the Harris corner detection algorithm. The results demonstrate the efficiency of the proposed approach compared with a typical two-dimensional implementation.

In the advent of autonomous machines, the need for real time video processing is becoming an increasingly important issue. Although technological advances have brought us closer to achieving this goal, they are often based on expensive and uniquely designed hardware solutions. It can be argued that as the complexity of image processing increases, it becomes more desirable to focus on portability and cost effective processing strategies. In this paper, we present a biologically inspired processing strategy that can be integrated with common, cost-effective image hardware. The results demonstrate that this approach can achieve a six-fold speedup, against a traditional image processing strategy, without any hardware modifications and a ten-fold speedup on adapted hardware. Alongside this, we present a novel type of processing that is used to detect video features in a space-time continuum. The results of this also demonstrate real-time processing potential and appear promising for motion focused tasks such as robot navigation.