Description:FIELD OF THE INVENTION

[0001] The present invention relates to a content-based image retrieval (CBIR) system that retrieves images based on visual content, ensuring accurate and relevant results. This enhances search efficiency by converting features for quick, reliable comparisons, thereby optimizing resource utilization and system performance.

BACKGROUND OF THE INVENTION

[0002] The Content-Based Image Retrieval (CBIR) is a technology designed to search and retrieve digital images based on their inherent visual characteristics like color, texture, and shape, rather than relying solely on textual metadata. By extracting and indexing these low-level visual features, CBIR allows users to query with an example image or sketch, comparing its features against a database to find visually similar results. This approach helps overcome the limitations of keyword-based searches, especially in large, untagged image collections, by directly analyzing image content to provide more relevant and contextually accurate retrieval for various applications, from medical imaging to e-commerce.

[0003] Traditional Content-Based Image Retrieval (CBIR) often fall short due to the "semantic gap", where basic visual features fail to capture complex human understanding of an image, leading to inaccurate and irrelevant results. These methods frequently rely on handcrafted features that are not only costly to develop but also struggle to generalize across diverse image collections. Consequently, they often exhibit poor efficiency and scalability when managing large databases, demanding significant computational resources for feature extraction and similarity matching. Moreover, the absence of adaptive learning or user feedback mechanisms in conventional CBIR systems means they remain static, unable to learn from user preferences or improve their retrieval accuracy over time, resulting in a less personalized and often frustrating user experience.

[0004] KR20220126845A discloses a system for providing a clothing pattern searching service using a big data-based CBIR. The system comprises: a user terminal which uploads product photos, receives a size input, and outputs pattern data of a clothing item, which is a subject within a product; and a search service providing server which includes a separation unit which extracts and separates a pattern of at least one portion of clothing from at least one clothing item; a big data conversion unit which maps and stores clothing data and pattern data to build a big data set; a search unit which inputs the product photos as queries using content-based image retrieval (CBIR) when the product photos and sizes are uploaded from the user terminal, and extracts clothing data with high similarity; and a transmission unit which extracts the pattern data mapped and stored in the searched clothing data, and transmits the pattern data to the user terminal. So, a user can easily obtain the desired pattern of clothing.

[0005] US10628736B2 discloses a content based image retrieval (CBIR) system and method is presented herein. The CBIR system generates a relatively short vector or array of data, referred to as a barcode, from an input image. The short vector or array data can be used to represent the content of the image for image retrieval purposes. The system obtains the image and applies a transform to the image to generate a plurality of image transform values. The system thresholds the plurality of image transform values to obtain compact image transform values. The system generates a barcode in accordance with the compact image transform values and representative of the image. The system may then transmit the barcode to a database for storage or draw the barcode on a display. The system may also compare barcodes to find and retrieve similar images associated with similar barcodes.

[0006] Conventionally, many systems are available in the market for (CBIR) system but existing systems often struggle with accurate visual content retrieval, leading to irrelevant results and inefficient searches. They typically lack the ability to quickly compare image features, hindering performance and resource optimization. Furthermore, they frequently fail to incorporate user feedback, resulting in less precise and unsatisfying image discovery.

[0007] In order to overcome the aforementioned drawbacks, there exists a need in the art to develop a system that offering accurate visual content retrieval, efficient comparisons, and user-feedback-driven accuracy, this system ensures precise and satisfying image discovery and a key market advantage.

OBJECTS OF THE INVENTION

[0008] The principal object of the present invention is to overcome the disadvantages of the prior art.

[0009] An object of the present invention is to develop a system that is capable of retrieves images based on their visual content, ensuring accurate and relevant results for users, thus enhancing the efficiency and effectiveness of image search and management.

[0010] Another object of the present invention is to develop a system that is capable of improve image retrieval efficiency by converting image features into a format that allows quick and reliable comparisons, therefore optimizing resource utilization and enhancing overall system performance.

[0011] Yet another object of the present invention is to develop a system that is capable of enhance retrieval accuracy by incorporating user feedback to adjust and optimize the system’s focus on relevant image characteristics, thus leading to increasingly precise and satisfying image discovery for users.

[0012] The foregoing and other objects, features, and advantages of the present invention will become readily apparent upon further review of the following detailed description of the preferred embodiment as illustrated in the accompanying drawings.

SUMMARY OF THE INVENTION

[0013] The present invention relates to a content-based image retrieval (CBIR) system improves image retrieval by converting features for quick, reliable comparisons, optimizing system performance. The system also enhances accuracy by integrating user feedback to refine focus on relevant characteristics, ensuring precise and satisfying image discovery.

[0014] According to an embodiment of the present invention, a content-based image retrieval (CBIR) system, comprising a U-NET Feature Extraction Unit connected to the Image Input Unit to extract fine-grained texture and color features from input images using a U-NET segmentation model, a feature vector conversion Unit connected to the U-NET Feature Extraction Unit to flatten extracted feature maps into high-dimensional vectors for efficient storage and comparison, a similarity computation Unit connected to the Feature Vector Conversion Unit that calculates cosine similarity scores between a query image vector and stored image vectors, a user feedback collection unit connected to the Display Output Unit to receive user feedback indicating whether retrieved images are relevant or irrelevant, a feature Re-Weighting Unit connected to the user feedback collection unit dynamically adjusts the importance of image features, the U-NET Feature Extraction Unit uses both the contracting and expanding paths to preserve spatial detail in texture and color segmentation, the User Feedback Collection Unit allows users to tag images as "relevant" or "irrelevant" during each search session and the Feature Re-Weighting Unit updates feature weights using positive and negative feedback with learning parameters α and β to emphasize user preferences.

[0015] While the invention has been described and shown with particular reference to the preferred embodiment, it will be apparent that variations might be possible that would fall within the scope of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

[0016] These and other features, aspects, and advantages of the present invention will become better understood with regard to the following description, appended claims, and accompanying drawings where:
Figure 1 illustrates a block diagram depicting workflow of a content-based image retrieval (CBIR) system.

DETAILED DESCRIPTION OF THE INVENTION

[0017] The following description includes the preferred best mode of one embodiment of the present invention. It will be clear from this description of the invention that the invention is not limited to these illustrated embodiments but that the invention also includes a variety of modifications and embodiments thereto. Therefore, the present description should be seen as illustrative and not limiting. While the invention is susceptible to various modifications and alternative constructions, it should be understood, that there is no intention to limit the invention to the specific form disclosed, but, on the contrary, the invention is to cover all modifications, alternative constructions, and equivalents falling within the spirit and scope of the invention as defined in the claims.

[0018] In any embodiment described herein, the open-ended terms "comprising," "comprises,” and the like (which are synonymous with "including," "having” and "characterized by") may be replaced by the respective partially closed phrases "consisting essentially of," consists essentially of," and the like or the respective closed phrases "consisting of," "consists of, the like.

[0019] As used herein, the singular forms “a,” “an,” and “the” designate both the singular and the plural, unless expressly stated to designate the singular only.

[0020] The present invention relates to a that retrieves images based on visual content, providing accurate and relevant results, thus enhancing image search efficiency and management. The system also boosts retrieval accuracy by integrating user feedback to optimize focus on relevant characteristics, leading to precise image discovery.

[0021] Referring to Figure 1, a block diagram depicting workflow of a content-based image retrieval (CBIR) system. The present invention relates to a Content-Based Image Retrieval (CBIR) system designed to efficiently retrieve visually similar images from a database by leveraging image processing, machine learning, and user feedback mechanisms. The system addresses the limitations of traditional image retrieval methods, which often rely on metadata or manual tagging, by utilizing a sophisticated architecture that extracts fine-grained features, computes similarities, and dynamically refines results based on user input. This invention provides an innovative, scalable, and adaptable solution for applications requiring precise image retrieval, such as digital archives, e-commerce, medical imaging, and multimedia search engines.

[0022] The CBIR system comprises five interconnected units, each performing a specialized function to ensure accurate and user-centric image retrieval. These units are: (1) a U-NET Feature Extraction Unit, (2) a Feature Vector Conversion Unit, (3) a Similarity Computation Unit, (4) a User Feedback Collection Unit, and (5) a Feature Re-Weighting Unit. Together, these components form a cohesive pipeline that processes input images, extracts meaningful features, compares them to a database, and iteratively improves retrieval performance based on user preferences.

[0023] The U-NET Feature Extraction Unit serves as the system's entry point, receiving raw images from the Image Input Unit. It employs a U-NET segmentation model, a convolutional neural network (CNN) architecture renowned for its effectiveness in image segmentation tasks. The U-NET model consists of a contracting path (encoder) that captures contextual information and an expanding path (decoder) that preserves spatial details, enabling the extraction of fine-grained texture and color features. By leveraging both paths, the unit ensures that intricate details, such as patterns, edges, and color gradients, are accurately captured, making it ideal for applications where visual fidelity is critical, such as medical imaging or art cataloging.

[0024] The Feature Vector Conversion Unit processes the feature maps generated by the U-NET model, flattening them into high-dimensional vectors suitable for efficient storage and comparison. This unit applies normalization techniques to ensure consistency across feature vectors, mitigating issues related to varying image sizes or intensities. The normalized vectors are stored in a structured database, optimized for rapid access during retrieval. This design enables the system to handle large-scale image databases efficiently, reducing computational overhead and improving retrieval speed.

[0025] The Similarity Computation Unit calculates the cosine similarity between the query image’s feature vector and the stored vectors in the database. Cosine similarity, a robust metric for measuring the angular distance between vectors, ensures that images with similar visual characteristics (e.g., texture, color, or patterns) are ranked higher. The unit outputs a ranked list of images, enabling users to quickly identify visually similar matches. This approach is particularly effective for applications requiring precise visual matching, such as product search in e-commerce or identifying similar artworks.

[0026] The User Feedback Collection Unit, connected to the Display Output Unit, allows users to provide feedback on the relevance of retrieved images. Users mark images as "relevant" or "irrelevant," providing the system with valuable insights into user preferences. This feedback mechanism ensures that the system adapts to individual or application-specific needs, enhancing retrieval accuracy over time.

[0027] The Feature Re-Weighting Unit dynamically adjusts the importance of image features based on user feedback. By analyzing patterns in the feedback, the unit re-weights feature (e.g., prioritizing texture over color for certain queries) to refine future retrievals. This iterative learning process ensures that the system continuously improves its performance, aligning results more closely with user expectations.

[0028] The system leverages deep learning, specifically the U-NET architecture, for feature extraction, combined with cosine similarity for efficient comparison. The integration of user feedback introduces an adaptive learning component, making the system highly flexible. The use of normalized feature vectors and a structured database ensures scalability, while the modular design allows for easy integration into various platforms.

[0029] This CBIR system is versatile, applicable in domains such as medical diagnostics (e.g., retrieving similar X-ray images), e-commerce (e.g., finding visually similar products), and digital archives (e.g., cataloging artworks). Its key advantages include high accuracy in feature extraction, fast retrieval through optimized vector storage, and adaptability via user feedback. The system overcomes the limitations of keyword-based search by focusing on visual content, offering a more intuitive and precise retrieval experience.

[0030] The present invention work best in the manner, where the Content-Based Image Retrieval (CBIR) efficiently retrieves visually similar images using a sophisticated, multi-unit architecture. The U-NET Feature Extraction Unit begins by processing raw images with the U-NET segmentation model, a convolutional neural network adept at capturing fine-grained texture, color, and spatial details. These extracted feature maps then move to the Feature Vector Conversion Unit, where they are flattened into high-dimensional, normalized vectors. These vectors are efficiently stored in a structured database, enabling rapid access and handling of large image collections. For retrieval, the Similarity Computation Unit calculates cosine similarity between the query image's feature vector and the stored vectors, producing a ranked list of visually similar images. Crucially, the User Feedback Collection Unit allows users to mark images as relevant or irrelevant, feeding insights to the Feature Re-Weighting Unit. This unit dynamically adjusts the importance of specific image features based on user preferences, ensuring the system continuously learns and refines its performance for more precise and satisfying image discovery.

[0031] Although the field of the invention has been described herein with limited reference to specific embodiments, this description is not meant to be construed in a limiting sense. Various modifications of the disclosed embodiments, as well as alternate embodiments of the invention, will become apparent to persons skilled in the art upon reference to the description of the invention. , Claims:1) A content-based image retrieval (CBIR) system, comprising:

i) a U-NET Feature Extraction Unit connected to the Image Input Unit to extract fine-grained texture and color features from input images using a U-NET segmentation model;
ii) a feature vector conversion Unit connected to the U-NET Feature Extraction Unit to flatten extracted feature maps into high-dimensional vectors for efficient storage and comparison;
iii) a similarity computation Unit connected to the Feature Vector Conversion Unit that calculates cosine similarity scores between a query image vector and stored image vectors to retrieve visually similar images;
iv) a user feedback collection unit connected to the Display Output Unit to receive user feedback indicating whether retrieved images are relevant or irrelevant; and
v) a feature Re-Weighting Unit connected to the user feedback collection unit dynamically adjusts the importance of image features based on user feedback to refine future retrievals.

2) The system as claimed in claim 1, wherein the U-NET Feature Extraction Unit uses both the contracting and expanding paths to preserve spatial detail in texture and color segmentation.

3) The system as claimed in claim 1, wherein the Feature Vector Conversion Unit normalizes and stores feature vectors in a structured database for fast access during retrieval.

4) The system as claimed in claim 1, wherein the Similarity Computation Unit uses cosine similarity to rank images based on their closeness to the query feature vector.

5) The system as claimed in claim 1, wherein the User Feedback Collection Unit allows users to tag images as "relevant" or "irrelevant" during each search session.

6) The system as claimed in claim 1, wherein the Feature Re-Weighting Unit updates feature weights using positive and negative feedback with learning parameters α and β to emphasize user preferences.