Matching Intensity for Image Visibility Graphs: A New Method to Extract Image Features

Abstract

Recently, the image visibility graphs (IVG) had introduced as simple algorithms by which images map into complex networks. However, current methods based on IVG use global statistical behaviors of the resulting graph to extract image features, which leads to loss of the local structural information of the image. To extract more informative image features by using the concept of IVG, we propose a new concept called matching intensity for image visibility graphs (MIIVG). The key idea of MIIVG is to separate the image into segments and represent the structural behavior of each with reference patterns and corresponding matching intensity. Theoretical analysis shows that the operation of MIIVG can be simplified to convolution operation and provides 256 convolution kernels with clear and apparent physical meaning, through which we can extract image features from multi-viewpoints and obtained more informative image features. Theoretical analysis and experiments demonstrate that MIIVG has a remarkable computing speed and is sufficiently stable against noise. Its high performance in image feature extraction we confirmed by two experiments. In keypoint matching experiments, MIIVG achieves a competitive result compared with SIFT. In texture classification experiments, compared with LBP, MIIVG is superior to LBP in calculation speed and classification effect. Compared with several current deep learning models, they all have the best feature extraction effect and very fast, but the features extracted by MIIVG are more concise. Also, MIIVG hardware requirements are lower, so it is easier to deploy. It is worth mentioning that MIIVG achieved 99.7% classification accuracy on the Multiband datasets, which is a state of the art performance on texture classification task of Multiband datasets and fully demonstrates the effectiveness of MIIVG.

An integrative method for COVID-19 patients’ classification from chest X-ray using deep learning network with image visibility graph as feature extractor

“…Also, we connect i and j in the HVG if ordering criterion is fulfilled: x k < inf(x i , x j ), k : i < k < j. The image horizontal visibility graph (IHVG) obeys the same set of conditions like IVG [39][40][41].…”

Visibility graph for time series prediction and image classification: a review

“…The texture classification based on image visibility graph (TCIVG) [ 84 ] considers the degree distribution of IVG and IHVG to classify images, and the best classification accuracy of this approach (91.4%) is obtained by applying quadratic discriminant in the IVG without lattice. Images are then divided into multiple segments by the matching intensity for image visibility graphs (MIIVG) [ 85 ], and the structure is described by the reference patterns and matching intensity. The best classification accuracy of this approach is 99.7%, and it is obtained by linear discriminant in MIIVG-1 and quadratic discriminant in MIIVG-2.…”

“…The image visibility graph can be further used to classify images into different categories by combining with several machine learning models. Compared with traditional artificial intelligence models, including VGG16, VGG19, and convolutional neural network (CNN), the best classification accuracy obtained by image visibility graph algorithms [ 83 – 85 ] is higher, illustrating the effectiveness of the visibility graph algorithm.…”