Image Retrieval by Fusion of Features from Pre-trained Deep Convolution Neural Networks

Bhandi, Vijayakumar; Devi, K. A. Sumithra

doi:10.1109/icatiece45860.2019.9063814

Cited by 16 publications

(10 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The proposed multi-similarity function-based CBIR model was developed in Python. Here, the performance of the proposed model was compared over the conventional models like VGG-16 [27], Inception v3 [28], and Xception [30] and heuristic approaches like PSO [31], GWO [32], DHOA [33], and ROA [26] in terms of F1-score, Precision and Recall. The evaluation was carried out on "two datasets" and analyzed the performance of retrieved images with various performance measures.…”

Section: Methodsmentioning

confidence: 99%

“…The features extracted from VGG-16 for both database images and query images are correspondingly known as 𝐹 𝑠 𝑉𝐺𝐺−16 and 𝑉 𝑠 𝑉𝐺𝐺−16 . Inception v3: Inception v3 is one of the networks of CNNs, which is especially utilized for extracting the features from the both query images and database images [28]. It has the benefit of factoring convolutions into various branches successively operating on space and channels.…”

Section: Trio-based Deep Feature Extractionmentioning

confidence: 99%

“…The VGG-16 [27] model improves accuracy mostly through increasing model depth. Inception [28] is capable of extracting multi-level features. Xception [30] substitutes regular Inception modules with depth-wise separable convolutions to further improve accuracy.…”

Section: Investigation On Deep Learning Techniquesmentioning

confidence: 99%

See 2 more Smart Citations

Enhanced Content-Based Image Retrieval with Trio-Deep Feature Extractors with Multi-Similarity Function

2022

IJIES

View full text Add to dashboard Cite

Introduction of the World Wide Web (WWW) and progressions in the field of multimedia and computer technology have enlarged the amount of image collections and databases including art galleries, digital libraries, and medical imageries that extend to millions of images. From these large-scale datasets, the retrieval procedure of images is carried out by classical approaches like chi square distance, colour histogram and text-based image retrieval, which however requires more time for getting the desired images. Thus, there is a need of proposing an effective retrieval system for images by handling these large numbers of images. For this reason, a new Content-Based Image Retrieval (CBIR) system has been implemented in this paper for retrieving the desired images of users from the collected images. Initially, different kinds of images like medical images, texture images, and environmental images etc., are collected from the standard image database. The deep feature extraction using Visual Geometry Group network (VGG-16), Inception v3, and Xception are used to get the features of all the images in the database. The parameters in these three deep learning techniques are tuned with an enhanced optimization algorithm of Modified Bypass-Rider Optimization Algorithm (MB-ROA). The features from VGG16, Inception and Xception are used for optimal weighted fused feature selection using enhanced ROA under training phase. During testing with query images, it undergoes deep feature extraction with the same set of techniques and, then fed into the optimal weighted fused feature selection to get the features of query image. Then, the multi-similarity function considering the cosine, Euclidean and Jaccard similarity are checked between the optimal features of database images and query image. The database images with the minimum similarity with the query image are retrieved from the database. The overall precision and recall to retrieve the top ten images related to the provided query image from the Corel dataset are 77.25% and 76.89%, respectively, while 70.15% and 70.26%, respectively from the VisTex dataset. The experimental findings show that the recommended model is effective at retrieving images from the database.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Trio-based Deep Feature Extractionmentioning

confidence: 99%

See 1 more Smart Citation

Enhanced Content-Based Image Retrieval with Trio-Deep Feature Extractors with Multi-Similarity Function

2022

IJIES

View full text Add to dashboard Cite

show abstract

“…Based on previous success in image retrieval using features extracted by neural networks [1,4,22,31], we further explore ResNet [15] as a feature extractor. As our data do not have labels, we extract features by a network pretrained on ImageNet [22] and build a BoW the same way as in the experiments using SIFT/SURF.…”

Section: S-cbir System Using Bowmentioning

confidence: 99%

Cross-Modality Sub-Image Retrieval using Contrastive Multimodal Image Representations

Breznik¹,

Wetzer²,

Lindblad³

et al. 2022

Preprint

View full text Add to dashboard Cite

Image Representations for Registration (CoMIRs) can be used to reduce the difficult task of image retrieval across modalities to a simpler one of image retrieval within the same modality.• We combine state-of-the-art contrastive representation learning with fast and robust feature extractors such as SURF to create a bag-of-words model for multimodal image retrieval.• Imposing rotation equivariance to the learned joint representations provides high performance of cross-modality retrieval.• The proposed approach outperforms bag-of-words built upon GAN-based image-to-image translations, as well as a recent reverse image search toolkit using shape descriptors, on a publicly available multimodal dataset of brightfield and second harmonic generation microscopy images.

show abstract

“…Various recent literatures use deep learning-based features for CBIR problem and have achieved considerable performance improvement than traditional feature extraction methods. [35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51] 1.…”

Section: Introduction 11 Related Workmentioning

confidence: 99%

Content‐based image retrieval for super‐resolutioned images using feature fusion: Deep learning and hand crafted

Pathak

Raju

2022

Concurrency and Computation

View full text Add to dashboard Cite

An innovative image retrieval agenda by concatenating deep learning features from GoogleNet and low‐level features from HSI and RGB color space is proposed in this article. Most of the CNN features suffer from loss of information due to image resize as a pre‐processing stage. To reduce this information loss super‐resolution technic is used for resizing images. An improved form of dot‐diffused block truncation coding is used for extracting RGB handcraft features. To discover the interdependencies between color and intensity component of an image, interchannel voting between hue, saturation, and intensity component is calculated as a color feature in HSI space. Histogram of orientated gradient (HOG) feature is used as shape feature. Five standard performance parameters, average precision rate, average recall rate, F‐Measure, Average Normalized Modified Retrieval Rank, and Total Minimum Retrieval Epoch, are applied on nine image datasets: Corel‐1K, Corel‐5K, Corel‐10K, VisTex, STex, ColorBrodatz and three subsets of ImageNet dataset for evaluation process of proposed method. For all dataset the best performance is achieved by the proposed method with respect to all performance parameters.

show abstract

Image Retrieval by Fusion of Features from Pre-trained Deep Convolution Neural Networks

Cited by 16 publications

References 30 publications

Enhanced Content-Based Image Retrieval with Trio-Deep Feature Extractors with Multi-Similarity Function

Enhanced Content-Based Image Retrieval with Trio-Deep Feature Extractors with Multi-Similarity Function

Cross-Modality Sub-Image Retrieval using Contrastive Multimodal Image Representations

Content‐based image retrieval for super‐resolutioned images using feature fusion: Deep learning and hand crafted

Contact Info

Product

Resources

About