Search

The expeditious proliferation of web services over the current web-cyberspace becomes a significant challenge for the web service classification. Recent reports suggest a deep learning algorithm namely, gated recurrent unit, performs well. But the performance of the GRU classifier hinges on structure of the GRU classifier namely, hidden neural network (HNN) and training data. Hence, it is significant to predict optimal HNN and training dataset. The recent studies revealed that genetic algorithm (GA) is most suited for this optimisation problem. However, GA is inefficient in terms of convergence rate and generation of prudent solution. For that, domain knowledge is incorporated in GA operator in order to accelerate convergence rate called as knowledge-based genetic algorithm (KBGA). The research work adopts KBGA to optimise the structure of GRU, so that the classifier outperforms the classification process. Extensive analysis and statistical hypothesis test have been done to show the efficacy of KBGA.

Retinal images are extensively used for the disclosure of retinal vascular disorders like diabetic retinopathy, glaucoma, age-related macular degeneration and optic neuritis. The analysis of a diagnostic retinal image using a computer is necessary to ease the optometrist for automating the load screening mechanism to identify these disorders. The primary step in technology aided diagnoses is optic disc segmentation and it is considered as an interesting search problem. The proposed methodology involves pre-processing of fundus images and optic disc localisation using Jaya algorithm. A new fitness function is proposed in order to improve the accuracy of the optic disc localisation. The efficiency of the proposed methodology is examined on different datasets such as DIARETDB1, DIARETDB0, CHASEDB1, DRIONS and DRIVE. The results infer that the proposed methodology has achieved a 99% of location accuracy and shows its superiority in localising the optic disc by comparing it with existing literature.

The design of quantisation table is viewed as an optimisation problem because the quantisation table produces the compression/quality trade-off in baseline joint photographic experts group algorithm. In this paper, efforts have been taken to reduce the computation time of the differential evolution (DE) algorithm by using the surrogate model. This paper applies a problem approximation surrogate model (PASM) to assist DE algorithms for optimising the quantisation table. It also analyses the performance of PASM in DE algorithm based on approximation error and evolutionary perspective. In addition, it confirms the results using statistical hypothesis tests. PASM is integrated in classical differential evolution and knowledge-based differential evolution algorithms. Different benchmark images are used to validate the PASM performance in DE algorithms for three target bits per pixel. The result shows that integrated PASM in DE algorithms reduces the computation time and guarantees the similar results as DE algorithms without a model.

Artificial intelligence (AI) allows the systems to observe from environments, perform certain functionalities and aims to maximise the probability of success in solving real world problems. With the technological enhancements and scientific growth, AI turns out to be a field of interest. Thus, it leads to the amplified focus on machine learning (ML) techniques. Machine learning (ML) is the most important data analysis methods which iteratively learn from the available data by using learning algorithms. The present survey provides the theoretical representation and basic methodologies of machine learning techniques like support vector machine (SVM), K-nearest neighbours (KNNs), decision tree, Bayesian networks, clustering, hidden Markov model (HMM) and neural networks. This survey paper provides the influence of machine learning techniques like clustering, SVM and ANN on image compression and attention to the existing scope for the image compression with machine learning.

Image quality/compression trade-off mainly depends on quantisation table used in JPEG scheme. Therefore, the generation of quantisation table is an optimisation problem. Even though recent reports reveal that classical differential evolution (CDE) is a promising algorithm to generate the optimal quantisation table, it is slow in convergence rate due to its weak local exploitation ability. This paper proposes knowledge-based differential evolution (KBDE) algorithm to search the optimal quantisation table for the target bits/pixel (bpp). KBDE incorporates the image characteristics and knowledge about image compressibility in CDE operators to accelerate the search. KBDE and CDE algorithms have been experimented on variety of images and an extensive performance analysis has been made between them, which reveal that KBDE accelerates the convergence rate of CDE without compromising on the quality of solution. Further, a statistical hypothesis test (t-test) confirms the result.