A novel image retrieval technique based on semi supervised clustering

Traditionally Content-Based Image Retrieval (CBIR) problems investigate the occurrence of images matching to a user-submitted query image or a sketch drawn by the user within a large image collection. However, there is often limited support for retrieving semantically similar images from large datab...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Multimedia tools and applications Jg. 80; H. 28-29; S. 35741 - 35769
Hauptverfasser: S, Nisha Chandran, Gangodkar, Durgaprasad
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.11.2021
Springer Nature B.V
Schlagworte:
1166: Advances of machine learning in data analytics and visual information processing
Clustering
Computer Communication Networks
Computer Science
Data Structures and Information Theory
Decision trees
Image management
Image retrieval
Matching
Multimedia Information Systems
Semantics
Similarity
Special Purpose and Application-Based Systems
Kullback -Leibler distance
Gaussian mixture modeling
Clustering
Decision trees
CBIR
Micro clusters
ISSN:1380-7501, 1573-7721
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Traditionally Content-Based Image Retrieval (CBIR) problems investigate the occurrence of images matching to a user-submitted query image or a sketch drawn by the user within a large image collection. However, there is often limited support for retrieving semantically similar images from large databases, matching the user’s perception. In this paper, we try to address this semantic gap problem in CBIR by performing a clustering-based retrieval. In the proposed approach we first perform a continuous probabilistic semi-supervised clustering to group similar images to form macro clusters. Macro clusters so formed, ensures class-wise similarity instead of semantic similarity. To retrieve the semantically matching images from these macro clusters formed, the CBIR method is adopted using a cluster within-cluster approach. The key idea is that the macro clusters formed during the initial phase of classification are further classified into micro clusters based on the decision tree approach. For retrieval, as the first step, the macro cluster matching to the user’s query is found. In the next step, to ensure semantic similarity the image is classified to the matching micro cluster. The proposed method is experimentally evaluated first on Wang database which contains complex and diverse images with varying fine details. Further, the experiments are repeated on the Ponce group database and Corel 5K database. The experimental results obtained demonstrate the effectiveness of the proposed approach.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1380-7501
1573-7721
DOI:10.1007/s11042-021-11542-3