Content uploaded by Simon Lawrence Harding
Author content
All content in this area was uploaded by Simon Lawrence Harding
Content may be subject to copyright.
A preview of the PDF is not available
Evolving Novel Image Features Using Genetic Programming-based Image Transforms
Abstract and Figures
In this paper, we use Genetic Programming (GP) to define a set of transforms on the space of greyscale images. The motivation is to allow an evolutionary algorithm means of transforming a set of image patterns into a more classifiable form. To this end, we introduce the notion of a transform-based evolvable feature (TEF), a moment value extracted from a GP-transformed image, used in a classification task. Unlike many previous approaches, the TEF allows the whole image space to be searched and augmented. TEFs are instantiated through Cartesian Genetic Programming, and applied to a medical image classification task, that of detecting muscular dystrophy-indicating inclusions in cell images. It is shown that the inclusion of a single TEF allows for significantly superior classification relative to predefined features alone.
Figures - uploaded by Simon Lawrence Harding
Author content
All figure content in this area was uploaded by Simon Lawrence Harding
Content may be subject to copyright.
... The works of [13], [14] fall in this category. In addition, there have been systems that use terminal sets containing as many variables as the number of pixels in an image patch of fixed size; for a recent example see [15]. ...
... A modular feed-forward architecture is reported in the work of [15]. It is defined by cascading a transformation layer, a pooling layer and a classification layer. ...
This book highlights various evolutionary algorithm techniques for various medical conditions and introduces medical applications of evolutionary computation for real-time diagnosis. Evolutionary Intelligence for Healthcare Applications presents how evolutionary intelligence can be used in smart healthcare systems involving big data analytics, mobile health, personalized medicine, and clinical trial data management. It focuses on emerging concepts and approaches and highlights various evolutionary algorithm techniques used for early disease diagnosis, prediction, and prognosis for medical conditions. The book also presents ethical issues and challenges that can occur within the healthcare system. Researchers, healthcare professionals, data scientists, systems engineers, students, programmers, clinicians, and policymakers will find this book of interest. © 2023 T. Ananth Kumar, R. Rajmohan, M. Pavithra, S. Balamurugan. All rights reserved.
Today, in the age of artificial intelligence and machine learning, Data mining and Image processing are two important platforms. GA and GP are value based and program based randomized searching tools respectively and these two are very much useful in the fields’ data mining and image processing for handling different issues. In this chapter, a review is made on ability of GA and GP in some applications of these two fields. Here, the selected subfields of data mining are market analysis, fraud detection, risk management, sports analysis, protein interaction, classification of data, drug discovery and feature construction. The similar in image processing are enhancement and segmentation of images, face recognition, photo mosaic generation, data embedding, image pattern classification, object detection and Graphics Processor Unit (GPU) development. The efficiencies of GA and GP in these particular applications are analyzed with corresponding parameters, comparing with other non-GA and non-GP approaches of the corresponding subfields.
This paper presents a Genetic Programming approach to feature extraction in the frame of the perceptron algorithm described in [1]. While feature extraction has the potential of increasing the accuracy of classification, fully exploring the huge space of possible combinations of the initial 45150 features would make the approach infeasible; Genetic Programming provides a proper way of tackling the search for relevant features. In turn, the extracted features are used to train an algorithm - One Side Class Perceptron - designed to minimize the number of false positives; accuracy is increased. In the experiments, the classifier using the extracted features was run on a dataset consisting of 358,144 files. The results show that our overall approach and implementation is fit for real-world malware detection.
Within the field of Computer Vision, change detection algorithms aim at automatically detecting significant changes occurring in a scene by analyzing the sequence of frames in a video stream. In this paper we investigate how state-of-the-art change detection algorithms can be combined and used to create a more robust algorithm leveraging their individual peculiarities. We exploited Genetic Programming (GP) to automatically select the best algorithms, combine them in different ways, and perform the most suitable post-processing operations on the outputs of the algorithms. In particular, algorithms’ combination and post-processing operations are achieved with unary, binary and n-ary functions embedded into the GP framework. Using different experimental settings for combining existing algorithms we obtained different GP solutions that we termed IUTIS (In Unity There Is Strength). These solutions are then compared against state-of-the-art change detection algorithms on the video sequences and ground truth annotations of the ChangeDetection. net (CDNET 2014) challenge. Results demonstrate that using GP, our solutions are able to outperform all the considered single state-of-the-art change detection algorithms, as well as other combination strategies. The performance of our algorithm are significantly different from those of the other state-of-the-art algorithms. This fact is supported by the statistical significance analysis conducted with the Friedman Test and Wilcoxon Rank Sum post-hoc tests.
Sample supervised search-centric image analysis has garnered research attention in various imaging disciplines concerned, to some extent, with identifying multitudes of target elements in an image. In a remote sensing image analysis context such a methodological philosophy might find value in a rapid mapping scenario or even as part of general land cover mapping approach. A user is not involved in the decisions of how to construct or tune image processing for the given information extraction problem, but is rather tasked to give a few examples of the elements of interest, similar to the concept of supervised classification. The information extraction method is automatically constructed, with the construction process guided by various notions of quality derived from comparing the limited user provided samples with generated results. Such a design philosophy holds substantial value for rapid or emergency response mapping endeavours, such as the Copernicus Emergency Management Services (EMS). Current land-cover mapping methodologies and technologies are not tailored to the needs of rapid mapping. The concept of sample supervised optimization based image analysis is examined, in which an efficient search algorithm is used, in various ways, to construct or tune a method. The application of such a general search-centric information extraction philosophy within the discipline of Geographic Object-based Image Analysis (GEOBIA) is reviewed, with a few discussions given on limitations of such a general approach. GEOBIA, as a sub-discipline of both geographic information science and remote sensing, denotes various methodological approaches to image analysis centred on the concept of semantic segmentation. Finally, considering developments in sample supervised image analysis from outside of the domain of remote sensing, a synthesis is presented on potential avenues for research on sample supervised search-centric methods within GEOBIA.
This chapter presents a review of some of the most representative work regarding techniques and applications of evolutionary algorithms in pattern recognition. Evolutionary algorithms are a set of metaheuristics inspired on Darwins “survival of the fittest” principle which are stochastic in nature. Evolutionary algorithms present several advantages over traditional search and classification techniques, since they require less domain-specific information, are easy to use and operate on a set of solutions (the so-called population). Such advantages have made them very popular within pattern recognition (as well as in other domains) as will be seen in the review of applications presented in this chapter.
This paper describes an approach to image thresholding that combines various multiscale and morphological features, including texture, shape and edge filtering, by using genetic program- ming, to detect the presence of landslides and their active sectors in change detected multitem- poral aerial images.
The evolution of image filters using Genetic Pro-gramming is a relatively unexplored task. This is most likely due to the high computational cost of evaluating the evolved programs. We use the parallel processors available on modern graphics cards to greatly increase the speed of evaluation. Previous papers in this area dealt with noise reduction and edge detection. Here we demonstrate that other more complicated processes can also be successfully evolved, and that we can "reverse engineer" the output from filters used in common graphics manipulation programs.
Genetic algorithms were used for feature selection and creation in
two pattern-classification problems. On a machine-version inspection
task, it was found that genetic algorithms performed no better than
conventional approaches to feature selection but required much more
computation. On a difficult artificial machine-vision task, genetic
algorithms were able to create new features (polynomial functions of the
original features) which dramatically reduced classification error
rates. Neural network and nearest-neighbor classifiers were unable to
provide such low error rates using only the original features
An automatic evolutionary search is applied to the problem of feature extraction in an OCR application. A performance measure based on feature independence is used to generate features which do not appear to suffer from peaking effects [17]. Features are extracted from a training set of 30 600 machine printed 34 class alphanumeric characters derived from British mail. Classification results on the training set and a test set of 10 200 characters are reported for an increasing number of features. A 1.01 percent forced decision error rate is obtained on the test data using 316 features. The hardware implementation should be cheap and fast to operate. The performance compares favorably with current low cost OCR page readers.
This paper presents a method for generating morphological template features for use in a word verification system. Word verification is a process of matching an image of a word against a list of candidate strings. The verification process assigns a confidence value to each string. The rank of the correct string is one measure of success. A character by character rank may also be obtained. The paper describes experiments using 20 samples in each of 57 character classes to automatically develop 60 template features for the word verification process. The generation process is guided by a measure of orthogonality between the features. Two slightly different versions of the orthogonality measure are compared. The resulting features are then trained on roughly 67 character samples to form the feature/character statistics. Results of the word verification process using the automatically generated features are compared with earlier results using a set of manually generated features. The principal result is that automatically generated template features perform better than the manually generated features.
This paper describes an attempt to make use of machine learning or self-organizing processes in the design of a pattern-recognition program. The program starts not only without any knowledge of specific patterns to be input, but also without any operators for processing inputs. Operators are generated and refined by the program itself as a function of the problem space and of its own successes and failures in dealing with the problem space. Not only does the program learn information about different patterns, it also learns or constructs, in part at least, a secondary code appropriate for the analysis of the particular set of patterns input to it.
In this chapter, we describe the original and most widely known form of Cartesian genetic programming (CGP). CGP encodes computational
structures, which we call ‘programs’ in the form of directed acyclic graphs. We refer to this as ‘classic’ CGP. However these
program may be computer programs, circuits, rules, or other specialized computational entities.