Attention based image retrieval

Download
index.pdf
2013
Özyer, Gülşah Tümüklü
Show full item record
173
views
222
downloads
This thesis proposes a content-based image retrieval (CBIR) system based on the human visual attention, called Attention-based Image Retrieval (ABIR). The proposed ABIR system handles CBIR problem from the perspective of human perception. An efficient visual attention model specific to CBIR problem derived from the computational visual attention model of Itti and Koch is suggested. ABIR system defines the CBIR system as an attention task, where query and images in the database are considered together to extract region of interests. The ABIR system consists of saliency map computing, region extraction, feature extraction and similarity matching steps using the saliency information. Bottom-up Normalization Algorithm, Top-down Normalization Algorithm and Top-down Feature Map Weighting Algorithm are proposed to compute the saliency maps. Bottom-up normalization and top-down normalization algorithms attack the normalization process of Itti-Koch model to compute saliency of images. Bottom-up normalization algorithm computes the normalization parameters from the all images in the dataset. On the other hand, top-down normalization algorithm normalizes the images in the dataset by using query image. Top-down feature map weighting algorithm combines the feature maps of an image in the dataset by using the query image. The features of salient regions are computed by using proposed the saliency-based feature integration algorithm and saliency-based feature selection algorithm. A saliency-based similarity matching algorithm ranks the images with respect to the query image. The proposed ABIR system is tested on STIM and SIVAL object datasets and high resolution airport images. The retrieval results are superior compared to the selected state of the art CBIR systems.
Image processing., Image processing, Content-based image retrieval., Visual perception., Pattern perception.
http://etd.lib.metu.edu.tr/upload/12616274/index.pdf
https://hdl.handle.net/11511/22878
Graduate School of Natural and Applied Sciences, Thesis

Suggestions

Superpixel based image sequence representation and motion estimation
İnce, Kutalmış Gökalp; Alatan, Abdullah Aydın; Demirekler, Mübeccel; Department of Electrical and Electronics Engineering (2017)
In this study a superpixel based representation of image sequences is proposed. For superpixel extraction, a novel gradient ascent approach, in which spatial and spectral statistics are utilized to obtain an optimal Bayesian classifier for pixel to superpixel label assignment, is proposed. Utilization of the spectral and spatial statistics reduce the dependency on user selected global parameters, while increasing the robustness and adaptability. Proposed Local Adaptive Superpixels (LASP) approach exploits he...
CONTENT BASED HYPERSPECTRAL IMAGE RETRIEVAL USING BAG OF ENDMEMBERS IMAGE DESCRIPTORS
Omruuzun, Fatih; Demir, Begum; Bruzzone, Lorenzo; Çetin, Yasemin (2016-08-24)
This paper proposes a novel system for fast and accurate content based retrieval of hyperspectral images. The proposed system aims at retrieving hyperspectral images that have both similar spectral characteristics associated with specific materials and fractional abundances to the query image. It consists of two modules. The first module characterizes the query and the target hyperspectral images in the archive by two descriptors: 1) a binary spectral descriptor representing spectral characteristics of dist...
Image generation using only a discriminator network with gradient norm penalty
Yeşilçimen, Cansu Cemre; Akbaş, Emre; Department of Computer Engineering (2022-9)
This thesis explores the idea of generating images using only a discriminator network by extending a previously proposed method (Tapli, 2021) in several ways. The base method works by iteratively updating the input image, which is pure noise at the beginning while increasing the discriminator's score. We extend the training procedure of the base network by adding the following new losses: (i) total variation, (ii) N-way classification (if labels are available), and (iii) gradient norm penalty on real exam...
Alignment of uncalibrated images for multi-view classification
Arık, Sercan Ömer; Vural, Elif; Frossard, Pascal (2011-12-29)
Efficient solutions for the classification of multi-view images can be built on graph-based algorithms when little information is known about the scene or cameras. Such methods typically require a pairwise similarity measure between images, where a common choice is the Euclidean distance. However, the accuracy of the Euclidean distance as a similarity measure is restricted to cases where images are captured from nearby viewpoints. In settings with large transformations and viewpoint changes, alignment of im...
Analysis of nanoparticle Transmission Electron Microscopy data using a public-domain image-processing program, Image
Woehrle, GH; Hutchison, JE; Özkar, Saim; Finke, RG (2006-01-01)
The need to easily and quickly count larger numbers of nanoparticles, in order to obtain statistically useful size and size-distribution data, is addressed via the use of a readily available, free, public-domain program for particle counting, NIH-Image (and 2 others derived from it, Scion Image and Image J), collectively referred to herein as Image. The best protocols that we have found useful for the use of Image are reported; both appropriate as well as problematic applications of Image are then illustrat...
Citation Formats
G. T. Özyer, “Attention based image retrieval,” Ph.D. - Doctoral Program, Middle East Technical University, 2013.
METU IIS - Integrated Information Service open@metu.edu.tr