{"created":"2023-05-15T15:29:22.324325+00:00","id":19054,"links":{},"metadata":{"_buckets":{"deposit":"7f142904-3ada-4432-a221-b326137bcd10"},"_deposit":{"created_by":15,"id":"19054","owners":[15],"pid":{"revision_id":0,"type":"depid","value":"19054"},"status":"published"},"_oai":{"id":"oai:sucra.repo.nii.ac.jp:00019054","sets":["94:429:431:432:966"]},"author_link":[],"item_113_alternative_title_1":{"attribute_name":"タイトル（別言語）","attribute_value_mlt":[{"subitem_alternative_title":"最適化技術に基づくフルリファレンスおよびノンリファレンス画像品質評価"}]},"item_113_biblio_info_9":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicIssueDates":{"bibliographicIssueDate":"2019","bibliographicIssueDateType":"Issued"}}]},"item_113_date_35":{"attribute_name":"作成日","attribute_value_mlt":[{"subitem_date_issued_datetime":"2020-07-21","subitem_date_issued_type":"Created"}]},"item_113_date_granted_20":{"attribute_name":"学位授与年月日","attribute_value_mlt":[{"subitem_dategranted":"2019-09-20"}]},"item_113_degree_grantor_22":{"attribute_name":"学位授与機関","attribute_value_mlt":[{"subitem_degreegrantor":[{"subitem_degreegrantor_name":"埼玉大学"}],"subitem_degreegrantor_identifier":[{"subitem_degreegrantor_identifier_name":"12401","subitem_degreegrantor_identifier_scheme":"kakenhi"}]}]},"item_113_degree_name_21":{"attribute_name":"学位名","attribute_value_mlt":[{"subitem_degreename":"博士(工学)"}]},"item_113_description_13":{"attribute_name":"形態","attribute_value_mlt":[{"subitem_description":"118 p.","subitem_description_type":"Other"}]},"item_113_description_23":{"attribute_name":"抄録","attribute_value_mlt":[{"subitem_description":"In this study, we proposed some objective image quality estimation methods for full-reference image quality assessment(FR-IQA ) and no-reference or blind image quality assessment (BIQA ). First, we proposed a new combination technique for\nfull-reference image quality assessment (FR - IQA ) by utilizing three better-recognized image quality assessment (IQA ) methods. For selecting the IQA methods, we firstly pick up the most appropriate IQA index for image quality databases and then add other two indices which have the most dissimilar features with the first index. Indeed, the combination of multiple IQA measures naturally emerges because of the shortcomings of single IQA indices for different types of distortions.\nOver the last several decades, IQA has been a topic of intense research in image processing field and it is the process of determining the level of accuracy of digital images. Nowadays, huge amount of images are daily produced for several purposes, for example, forecasting weather, finding diseases and monitoring criminals. For these reasons, it is very importance to keep the quality of such images at an acceptable visual level at the end-users after the production and transmission. Furthermore, accurate measurement of the image quality is an important step in many image-based applications. To achieve this goal, effective IQA algorithms are necessary and have recently become a very hot research topic.\nBasically, there are two types of image quality assessment called subjective image quality assessment and objective image quality assessment. Subjective Image Quality Assessment is the most reliable way to evaluate the visual quality of digital images perceived by the human observers. In practice, however, subjective image quality assessment is time consuming and very expensive. Thus, it is often used to construct image quality datasets and provide the ground-truth mean opinion scores (MOS ) for evaluating objective quality measures. Objective image quality assessment is automatically estimating the quality of images by algorithms instead of humans using MOS provided by human observers and it is more handy than the subjective IQA . To carry out this requirement, many researchers proposed various not only single but also combination IQA methods in recent times. However, all existing single and combination IQA methods still have some shortcomings to be able to get the highest performance for full-reference IQA . Therefore, we consider a simple and robust combination method that are suitable for all image databases. In our combination, we firstly pick up the most correlated IQA method for all types of distortions by applying the algorithm that is used to select the most appropriate method for combination. After choosing the first combined IQA method, we choose the one which has the biggest index ranking difference with the first one as the second combined IQA method, since it has the most different characteristics comparing to the first chosen combined IQA method. Following the same way, we decide the third one. After selecting the most appropriate methods, we combine the three methods by employing the weighting factors, exponentaited coefficients and constant values. Then, we optimize these parameter values by using the Particle Swarm Optimization (PSO ). Experimental results verified that the proposed method gives the best performance for various databases and outperforms other state-of-the-art not only traditional single methods but also previous combination methods.\nOn the other hand, it is very difficult to get the information of reference images for image quality estimation in reality. This gives a motivation to consider blind image quality assessment (BIQA ) methods which are able to measure the quality of distorted images without referencing the original images. Therefore, many researchers develop numerous BIQA approaches using Natural Scene Statistics (NSS ) based features. In most NSS based BIQA methods, features are extracted by the wavelet transform and they are usually very slow due to the use of computationally expensive image transformations. Thus, more recent techniques promote extracting features from the spatial domain, which leads to a significant reduction in computation time. However, all existing BIQA methods have still restrictions to get the highest performance. To overcome the restrictions, we consider to construct a very simple and robust end-to-end learning mechanism using convolutional neural network (CNN ). One of CNN s advantages is that it can take raw images as input and incorporate feature learning into the training process. Thus, in our work, we take distorted images labelled with Mean Opinion Score (MOS ) as inputs and output the related score for each image. Experimental results demonstrated that our proposed method outperforms other state-of-the-art ones.","subitem_description_type":"Abstract"}]},"item_113_description_24":{"attribute_name":"目次","attribute_value_mlt":[{"subitem_description":"Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13\nAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14\n1 Introduction 16\n1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16\n1.1.1 Full-reference Image Quality Assessment (FR - IQA syes) . 17\n1.1.2 No-reference Image Quality Assessment (NR - IQA syes) . 19\n1.2 Problem Statements and Objectives . . . . . . . . . . . . . . . . . . 22\n1.3 Overviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23\n1.4 Organization of the thesis . . . . . . . . . . . . . . . . . . . . . . . 26\n2 Combination of Dissimilar Feature-Scores for Image Quality Assessment Using Particle Swarm Optimization Algorithm 27\n2.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27\n2.2 Structural Similarity (SSIM) Measure . . . . . . . . . . . . . . . . . 30\n2.3 Multi-scale structural similarity (MS-SSIM) Measure . . . . . . . . 32\n2.4 Most apparent distortion (MAD) Measure . . . . . . . . . . . . . . 33\n2.5 Feature similarity (FSIM) Measure . . . . . . . . . . . . . . . . . . 38\n2.6 Feature similarity index for color images (FSIMC) . . . . . . . . . . 40\n2.7 Combined Full-Reference Image Quality Metric (CQM) . . . . . . . 41\n2.8 Extended Hybrid Image Similarity (EHIS) . . . . . . . . . . . . . . 42\n2.9 Exponentiated Combination of Two Scores for Image Quality Assessment (2SCM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42\n2.10 Optimized Three Scores Combination for Image Quality Assessment (3SCM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43\n2.11 Particle Swarm Optimization (PSO) Algorithm . . . . . . . . . . . 44\n2.12 Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46\n2.13 Experimental Results and Discussion . . . . . . . . . . . . . . . . . 48\n2.13.1 Image Databases . . . . . . . . . . . . . . . . . . . . . . . . 48\n2.13.2 Cross Validation . . . . . . . . . . . . . . . . . . . . . . . . 60\n2.13.3 Parameter Optimization . . . . . . . . . . . . . . . . . . . . 61\n2.13.4 Performance Evaluation . . . . . . . . . . . . . . . . . . . . 62\n2.13.5 Experimental Results . . . . . . . . . . . . . . . . . . . . . . 62\n2.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69\n3 Convolutional Neural Network for Blind Image Quality Assessment 77\n3.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77\n3.1.1 Blind or Referenceless Image Spatial Quality Evaluator (BRISQUE) 78\n3.1.2 Codebook Representation for No-Reference Image Assessment (CORNIA) . . . . . . . . . . . . . . . . . . . . . . . . 80\n3.1.3 High Order Statistics Aggregation (HOSA) . . . . . . . . . . 81\n3.1.4 Learning from Rankings for NR-IQA and Fine-tuning (RankIQA+FT) 82\n3.2 Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . 83\n3.2.1 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . 84\n3.2.2 Pooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85\n3.2.3 Batch Normalization . . . . . . . . . . . . . . . . . . . . . . 85\n3.2.4 Activation functions . . . . . . . . . . . . . . . . . . . . . . 86\n3.3 Proposed Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87\n3.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90\n3.4.1 NNABLA . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91\n3.5 Experimental Results and Discussion . . . . . . . . . . . . . . . . . 93\n3.5.1 Image Databases . . . . . . . . . . . . . . . . . . . . . . . . 93\n3.5.2 Training and Testing . . . . . . . . . . . . . . . . . . . . . . 96\n3.5.3 Performance Evaluation . . . . . . . . . . . . . . . . . . . . 96\n3.5.4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . 96\n3.5.5 Cross validation . . . . . . . . . . . . . . . . . . . . . . . . . 105\n3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106\n4 Conclusion and Future Work 107\n4.0.1 Summary of the Research . . . . . . . . . . . . . . . . . . . 107\n4.0.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 108","subitem_description_type":"Other"}]},"item_113_description_25":{"attribute_name":"注記","attribute_value_mlt":[{"subitem_description":"指導教員 : 島村徹也","subitem_description_type":"Other"}]},"item_113_description_33":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"subitem_description":"text","subitem_description_type":"Other"}]},"item_113_description_34":{"attribute_name":"フォーマット","attribute_value_mlt":[{"subitem_description":"application/pdf","subitem_description_type":"Other"}]},"item_113_dissertation_number_19":{"attribute_name":"学位授与番号","attribute_value_mlt":[{"subitem_dissertationnumber":"甲第1142号"}]},"item_113_identifier_registration":{"attribute_name":"ID登録","attribute_value_mlt":[{"subitem_identifier_reg_text":"10.24561/00019023","subitem_identifier_reg_type":"JaLC"}]},"item_113_publisher_11":{"attribute_name":"出版者名","attribute_value_mlt":[{"subitem_publisher":"埼玉大学大学院理工学研究科"}]},"item_113_publisher_12":{"attribute_name":"出版者名（別言語）","attribute_value_mlt":[{"subitem_publisher":"Graduate School of Science and Engineering, Saitama University"}]},"item_113_record_name_8":{"attribute_name":"書誌","attribute_value_mlt":[{"subitem_record_name":"博士論文（埼玉大学大学院理工学研究科（博士後期課程））"}]},"item_113_text_31":{"attribute_name":"版","attribute_value_mlt":[{"subitem_text_value":"[出版社版]"}]},"item_113_text_36":{"attribute_name":"アイテムID","attribute_value_mlt":[{"subitem_text_value":"GD0001148"}]},"item_113_text_4":{"attribute_name":"著者 所属","attribute_value_mlt":[{"subitem_text_value":"埼玉大学大学院理工学研究科（博士後期課程）理工学専攻"}]},"item_113_text_5":{"attribute_name":"著者 所属（別言語）","attribute_value_mlt":[{"subitem_text_value":"Graduate School of Science and Engineering, Saitama University"}]},"item_113_version_type_32":{"attribute_name":"著者版フラグ","attribute_value_mlt":[{"subitem_version_resource":"http://purl.org/coar/version/c_970fb48d4fbd8a85","subitem_version_type":"VoR"}]},"item_access_right":{"attribute_name":"アクセス権","attribute_value_mlt":[{"subitem_access_right":"open access","subitem_access_right_uri":"http://purl.org/coar/access_right/c_abf2"}]},"item_creator":{"attribute_name":"著者","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"YADANAR, KHAING","creatorNameLang":"en"},{"creatorName":"ヤダナ, カイ","creatorNameLang":"ja-Kana"}]}]},"item_files":{"attribute_name":"ファイル情報","attribute_type":"file","attribute_value_mlt":[{"accessrole":"open_date","date":[{"dateType":"Available","dateValue":"2020-07-21"}],"displaytype":"detail","filename":"GD0001148.pdf","filesize":[{"value":"8.3 MB"}],"format":"application/pdf","licensetype":"license_note","mimetype":"application/pdf","url":{"label":"GD0001148.pdf","objectType":"fulltext","url":"https://sucra.repo.nii.ac.jp/record/19054/files/GD0001148.pdf"},"version_id":"477d9187-3d00-44c7-b5a8-101571e99525"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"eng"}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourcetype":"doctoral thesis","resourceuri":"http://purl.org/coar/resource_type/c_db06"}]},"item_title":"Full-Reference and Non-Reference Image Quality Assessment Based on Optimization Technique","item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"Full-Reference and Non-Reference Image Quality Assessment Based on Optimization Technique","subitem_title_language":"en"}]},"item_type_id":"113","owner":"15","path":["966"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2020-07-21"},"publish_date":"2020-07-21","publish_status":"0","recid":"19054","relation_version_is_last":true,"title":["Full-Reference and Non-Reference Image Quality Assessment Based on Optimization Technique"],"weko_creator_id":"15","weko_shared_id":-1},"updated":"2023-06-23T01:54:12.131598+00:00"}