Font Size: a A A

Research On Metamorphic Testing Technology For Image Recognition Software

Posted on:2022-10-11Degree:MasterType:Thesis
Country:ChinaCandidate:J J JiangFull Text:PDF
GTID:2518306575961869Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Image recognition is a key research area of image processing.With the application of more and more widely,how to test the robustness of image recognition software has become a concern.In traditional testing,testers need to compare the expected output of the program with the actual observed output.However,in some application fields of image recognition software,the process of finding the expected output and verification is very difficult and error-prone,which is called Oracle problem.Under the influence of Oracle problem,there is a lack of testing technology suitable for image recognition system,this paper studies the test technology used in the image recognition software,and provides support for the test and evaluation of this kind of information system.The main work of this paper is as follows:1)Proposing a metamorphic testing framework for image recognition softwareBased on the idea of metamorphic testing,a set of image recognition system testing framework based on metamorphic testing is proposed according to the specific characteristics of the image recognition system.It clarifies the specific steps of the testing process,covers the full life of metamorphic testing,and plays a guiding role in the content of this paper.2)Proposing a Proposing relation construction method of image domainIn order to apply the metamorphic test to the image recognition software,the existing basic metamorphic relations in the image-oriented Field are sorted out and classified,the metamorphic relation construction method based on the generative counter network is proposed to establish the metamorphic relation library,and proposed an integrated transformation relation construction methods.3)Proposing key measurement index and automatic determination methodIn view of the lack of metamorphic test evaluation index,the key measurement index and calculation model of image recognition software are proposed,the automatic result determination method is proposed from the two application directions of image classification and target detection respectively according to the characteristics of image recognition software.4)Proposing the composite method of metamorphic relationsFor improving the test efficiency,we studied the compound strategy of metamorphic relationship,and constructed the new metamorphic relationship systematically through the compound method,and the compound strategy suitable For image recognition software is explored through experiments.5)Developing a prototype tools of metamorphic testingImplemnted a prototype testing tool according to the proposed metamorphic testing framework,showing the practical application value of the research results in this paper.
Keywords/Search Tags:Metamorphic testing, Image recognition, Generative adversarial networks, Metamorphic relation composition, Key metrics
PDF Full Text Request
Related items