China
Africa
Explore chapters and articles related to this topic
Introduction
Published in Neil Collings, Fourier Optics in Image Processing, 2018
Template matching is the comparison of a number of stored patterns (templates) with the image until the template is located within the image and its positional coordinates are retrieved. This is a successful technique for retrieving the pattern in a noisy image when there is an exact match between the stored pattern and the pattern in the image. However, due to the variability of a given image (for example, the pose, illumination, and expression of a face), template matching using the whole face as a template is less successful. Smaller size templates can be used more effectively. For example, the eyes on the face can be represented by two black circles separated by a defined distance. This is a template for the eyes that can be rotated by varying angles and searched for in the image. This will allow detection of the frontal face at varying angular pose.
Ibn al-Haytham: Founder of Physiological Optics?
Published in Azzedine Boudrioua, Roshdi Rashed, Vasudevan Lakshminarayanan, Light-Based Science: Technology and Sustainable Development, 2017
Note that al-Haytham gives an algorithmic view – the ‘soul’ is trained with known objects which it then categorises and forms templates. Perception of the image is achieved by studying the similarities between the image and the stored template. This template matching is a very common technique in pattern recognition and computer vision (e.g., Perveen et al., 2013). However, al-Haytham emphasises that it is not just a simple geometric template matching, but takes into account all known knowledge and properties of the object. He discusses how the ‘soul’ categorises a rose from a general flower: when sight perceives roses in bloom in some garden it will immediately perceive that these visible objects are roses on account of the particular color of the roses in addition to their being in a garden, before perceiving the round shape of their petals or their arrangement, and before perceiving all properties that constitute the form of roses. And if the roses resemble some other flowers, sight will perceive them in any case to be flowers and not leaves of trees or [other] plants (Sabra, p. 219).
The RIMcomb research project: Towards the application of building information modeling in Railway Equipment Engineering
Published in Jan Karlshøj, Raimar Scherer, eWork and eBusiness in Architecture, Engineering and Construction, 2018
S. Vilgertshofer, D. Stoitchkov, S. Esser, A. Borrmann, S. Muhič, T. Winkelbauer
Template Matching is a well-known method for the searching of a template image in a larger image. This is made by sliding the template image over the input (larger) image and comparing them at every position. The result of this method is a grayscale image with a size of (W−w+1, H−h+1), where W and H are the width and the height of the input image, w and h are the width and the height of the template image. We investigated different comparison methods for each one of which there is a normalized version (Kaehler and Bradski, 2016).
A Vehicle License Plate Detection and Recognition Method Using Log Gabor Features and Convolutional Neural Networks
Published in Cybernetics and Systems, 2023
Ahmed Zaafouri, Mounir Sayadi, Wei Wu
The character recognition algorithms are mainly based on learning-based and template matching approaches. Neural networks and deep learning-based methods are the most widely used in literature: Ktata, Benzarti, and Amiri (2013), Becerikli et al. (2007), Norul et al. (2006). Recently, deep learning neural networks give successful results for character recognition: Li, Wang, and Shen (2017), Li et al. (2018). Moreover, Montazzolli and Jung (2017) propose a CNN to segment and recognize the characters within a cropped LP. Bulan et al. (2017) developed an approach that achieved a high accuracy in LP recognition. Menotti et al. (2014) proposed the use of random CNNs to extract features for character recognition, achieving a significantly better performance. However, directly training the deep models for the LPR task does not achieves well enough performance Jing et al. (2019). Due to the low accuracy of these methods in the recognition phase that detects and recognize LPs as a sequence, we proceed in our work as usual approaches using the two fundamental phases: characters extraction and recognition.
High-speed vision measurement of vibration based on an improved ZNSSD template matching algorithm
Published in Systems Science & Control Engineering, 2022
Jian Luo, Bingyou Liu, Pan Yang, Xuan Fan
However, traditional image processing technologies used in high-speed cameras not only have high requirements for motion tracking algorithm, these technologies are also complex and troublesome (see Aqel et al., 2016) The standard method for tracking features between two images is through template matching in computer vision. The basic template matching algorithm calculates each position through a distortion function, which then measures the similarity between the template and the image. The area-based algorithm involves calculating a correlation function at each location of the search image in a raster scan manner. Several area-based methods have been introduced over the years, including mean absolute difference, product cross correlation (CC), sequential similarity detection algorithm, sum of absolute difference (SAD), and sum of variance (SSD) similarity measurement (see Sahani et al., 2011). After considering template matching, due to its better robustness, the zero-mean normalization sum of squared differences (ZNSSD) algorithm is often used in similarity measurements (see Bing et al., 2010; Pangercic et al., 2008). Compared with other algorithms, ZNSSD is less sensitive to the linear change of the illumination amplitude in the two comparison images, making it suitable for actual measurement.
An Illumination Pre-processing Method Using the Enhanced Energy of Discrete Wavelet Transform for Face Recognition
Published in IETE Journal of Research, 2020
A. Thamizharasi, Jayasudha J. S.
The geometric methods uses the size measures of eyes, nose, mouth etc. and its representation is economic. But, these methods were sensitive to measurement process. Template matching is more reliable than geometric method, but it is affected by variations of light source, pose and scale changes. The computation cost of graph models are high. Principal Component Analysis (PCA) are better in representation while, it smears the classes together and could not discriminate among the classes. Linear Discriminant Analysis (LDA) increases the discrimination among different classes, but the neighbourhood relationships between pixels can be removed. LDA performs well in controlled environments. But, the illumination variations, changes in facial expressions and pose changes affects the recognition accuracy [1].