Ant Lion Optimizer for Texture Classification: A Moving Convolutional Mask

Abstract

Texture classification is an important issue for a number of applications in machine vision, which could be addressed through learning texture features by using a convolutional mask. However, the traditional convolutional mask has a single orientation, and the learning ability is limited. In addition, the optimization process often falls into the local optima, and the discrimination capacity of the learned mask is unsatisfactory. Thus, a novel moving convolutional mask is presented to enhance the discrimination of specific texture features in the proposed approach. Furthermore, how to achieve the satisfactory convolutional mask is considered as a combinatorial optimization problem and acquired by maximizing the texture energy by using ant lion optimizer (ALO). The proposed approach was tested on some public images, and the results were compared with those of the state-of-the-art approaches. The experimental results showed that ALO has strong optimization ability, and the proposed method is robust, adaptive and superior to the improved grey level co-occurrence matrix (GLCM), directional statistical Gabor filter and Tuned convolutional mask in terms of the fitness value and classification accuracy which has reached 27 and 91%, respectively, for all images. INDEX TERMS Moving convolutional mask, texture classification, feature extraction, ant lion optimizer.

Ant Lion Optimization: Variants, Hybrids, and Applications

“…To detect the change in natural environment, texture can be considered as a significant characteristics. Wang et al [170] used ALO to acheive the satisfactory convolutional mask by considering it as a combinatorial optimization. Their results prove the ability of ALO in and classification accuracy which has reached 91% and fitness value with 27%.…”

Prediction model for recurrence of hepatocellular carcinoma after resection by using neighbor2vec based algorithms

“…This is because ML methods can help find and recognize the patterns and relationships of patients' treatments and risk factors from complex data sets. These discovered information can be used to effectively predict the future trending results of cancer types (Ji et al, 2016; Wang, Wang, Ye, & Yang, 2019). Many research results show that the prediction model based on ML can significantly improve the accuracy of cancer diagnosis and prediction of prognosis.…”

Hyperspectral Image Denoising With Dual Deep CNN

“…Another popular point of view is joint utilization of spatialspectral features [28], [44]- [46]. Spatial-spectral methods can often achieve relatively better results, but these methods have to deal directly with high-dimension data [51]- [53]. In [47], it is pointed out that these methods are insufficient to satisfy many complex situations, because the strength of the noise in the spatial and spectral domains is unequal.…”