Description:FIELD OF THE INVENTION

[0001] The present invention relates to a system for retrieving images based on visual content that is capable of retrieve of the images in the higher accuracy by extraction the features on both high levels and low levels.

BACKGROUND OF THE INVENTION

[0002] Retrieving images based on visual content involves utilizing advanced image recognition and computer vision techniques to analyze and understand the features within an image, such as shapes, colors, textures, and objects. By extracting these visual attributes, systems can match or find similar images in large databases, enabling efficient content-based image retrieval (CBIR). This process often employs deep learning models like convolutional neural networks (CNNs) to generate feature vectors that represent the images' semantic content, facilitating accurate and rapid search results based solely on visual similarity rather than textual metadata.

[0003] Traditionally, retrieving images based on visual content relied on manual or rule-based methods such as metadata tagging, keyword annotations, or rudimentary feature extraction like color histograms and edge detection. While these approaches allowed some level of content-based search, they had notable drawbacks, including dependence on accurate and consistent manual annotations, which are time-consuming and prone to errors or subjectivity. Additionally, such methods often struggled to capture complex semantics or high-level concepts within images, leading to limited retrieval accuracy and relevance. They also faced scalability issues, as managing and searching through large image datasets became increasingly challenging without automated, robust feature extraction and indexing techniques, resulting in inefficiencies and reduced effectiveness in real-world applications.

[0004] WO2015032585A1 discloses a method and non-transitory computer readable medium for content based image retrieval. The method includes selecting a query image, segmenting the selected query image by applying a segmentation technique, extracting features from the segmented query image by determining at least two feature descriptors, including color feature descriptors and texture feature descriptors, and determining a similarity of the query image to a plurality of images included in a database using the determined at least two feature descriptors of the segmented query image, features being extracted from each of the plurality of images included in the database by determining the at least two feature descriptors, the color feature descriptors and the texture feature descriptors including a simultaneous combination of different color spaces, and global and local statistical measurements being carried out on the simultaneous combination of the different color spaces.

[0005] US20160078057A1 discloses a method and non-transitory computer readable medium for content based image retrieval. The method includes selecting a query image, segmenting the selected query image by applying a segmentation technique, extracting features from the segmented query image by determining at least two feature descriptors, including color feature descriptors and texture feature descriptors, and determining a similarity of the query image to a plurality of images included in a database using the determined at least two feature descriptors of the segmented query image, features being extracted from each of the plurality of images included in the database by determining the at least two feature descriptors, the color feature descriptors and the texture feature descriptors including a simultaneous combination of different color spaces, and global and local statistical measurements being carried out on the simultaneous combination of the different color spaces.

[0006] Conventionally, many devices have been developed to retrieve images based on visual content but these devices lack the retrieval of visually similar images precisely and adapting to the user preferences.

[0007] In order to overcome the aforementioned drawbacks, there exists a need in the art to develop a device that is capable of high accuracy retrieval of the images works efficiently even with a large amount of image data, makes the system more computational and storage efficient.

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 device that is capable of significantly enhance the accuracy of image retrieval in a variety of image database.

[0010] Another object of the present invention is to develop a device that is capable of preserving spatial hierarchies and capturing both low-level and high-level features simultaneously.

[0011] Another object of the present invention is to develop a device that is capable of capturing more complex and detailed features of images, making it well suited for domains where fine grained image characteristics are essential.

[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 system for retrieving images based on visual content that is capable of capturing more complex and detailed features of images, making it well suited for domains where fine grained image characteristics are essential meanwhile preserving spatial hierarchies and capturing both low-level and high-level features simultaneously.

[0014] According to an embodiment of the present invention, a system for retrieving images based on visual content comprising a user-interface installed in a computing unit wirelessly linked with the system via a communication module, that is accessed by a user to uploaded input data, that is transmitted to a pre-Processing Unit configured to input images, normalizes and augments images to prepare for feature extraction, a U Net Feature Extraction Unit linked with Pre-Processing Unit, to extract low-level and high-level features using a U Net architecture, wherein U Net Feature Extraction has an encoder-decoder architecture that includes skip connections to preserve spatial context during feature extraction, a feature re-weighting unit inked to the U Net Feature Extraction Unit, dynamically adjusts feature weights based on query relevance, the feature re-weighting unit uses a query-aware protocol to prioritize texture or shape features based on the query image content.

[0015] According to another embodiment of the present invention, the device further comprises of a similarity measurement unit connected to the feature re-weighting Unit, calculates cosine similarity to rank images, the similarity measurement unit cosine similarity scores are used to rank images in descending order of relevance to the query, a retrieval and ranking unit linked to the similarity measurement unit, retrieves and ranks images based on similarity scores, an display Unit linked to the Retrieval and Ranking Unit, displays ranked images to the user, the display unit provides a visualization dashboard displaying precision, recall, and F1-score metrics for retrieval performance, system further includes a user feedback assembly connected to the feature re-weighting unit to refine feature weights based on user input.

[0016] 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

[0017] 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 flowchart depicting the workflow of the system for retrieving images based on visual content.

DETAILED DESCRIPTION OF THE INVENTION

[0018] 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.

[0019] 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.

[0020] 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.

[0021] The present invention relates to a system for retrieving images based on visual content that is capable of significantly enhance the accuracy of image retrieval in a variety of image database while preserving spatial hierarchies and capturing both low-level and high-level features simultaneously.

[0022] Referring to Figure 1, a flowchart depicting the workflow of the system for retrieving images based on visual content.

[0023] The device disclosed herein includes a user-interface installed in a computing unit that is wirelessly linked with the system. A user to upload input data accesses the user-interface. The interface installed in a computing unit that is wirelessly linked to the system that enable seamless communication and data transfer. When the user accesses this interface to upload input data, they interact via a device such as a smartphone, tablet, or wireless-enabled terminal that communicates over protocols like Wi-Fi, Bluetooth, or cellular networks. The user’s input is captured by the interface such as touchscreen, keyboard and transmitted wirelessly through the network to the system’s server or processing unit. On the backend, the data is received by middleware that authenticates the user, processes the input, and routes it to the appropriate system component for storage. The entire process relies on standardized communication protocols, real-time data encoding, and security measures like encryption to ensure data integrity and privacy during transmission, enabling users to upload data remotely and interact with the system efficiently.

[0024] The input data is then transmitted to a pre-processing unit to extract low-level and high-level features using a U-Net architecture. The U Net Feature Extraction has an encoder-decoder architecture that includes skip connections to preserve spatial context during feature extraction. The pre-processing unit utilizing a U-Net architecture functions by first passing input data such as images through an encoder path that progressively captures low-level features (edges, textures) via successive convolutional and pooling layers, reducing spatial resolution while enriching feature representation. The encoder's output at each level is stored as skip connections, which are later concatenated with corresponding decoder layers to retain spatial details. The decoder path then performs up sampling through transposed convolutions, reconstructing the feature maps back to the original resolution while integrating high-level semantic information. This process enables the network to effectively extract both low-level features (fine details) and high-level features (contextual understanding) by combining coarse, abstract representations with preserved spatial context via skip connections, ensuring precise localization and recognition during feature extraction.

[0025] The U-Net Feature Extraction Unit is linked with a feature re-weighting unit. The re- weighting unit dynamically adjusts feature weights based on query relevance by using a query-aware protocol to prioritize texture or shape features based on the query image content. The re-weighting unit functions through a query-aware protocol that assesses the relevance of different features such as texture or shape based on the content of a given query image. It employs a relevance scoring module, using attention metrics, to analyze the query features and generate a set of adaptive weights. These weights are then used to recalibrate the feature maps extracted from the target image, emphasizing the features most pertinent to the query’s context. This process involves element-wise multiplication or gating arrangements that amplify relevant features (e.g., textures in one query, shapes in another) while suppressing less relevant information, enabling the system to focus on query-specific details dynamically. Consequently, the re-weighting unit enhances the model’s ability to prioritize features that improve task performance, such as matching textures or shapes aligned with the query image content.

[0026] With the feature re-weighting Unit, a similarity measurement unit is connected that calculates cosine similarity to rank images in descending order of relevance to the query. The similarity measurement unit operates by computing the cosine similarity between feature vectors extracted from the query image and candidate images, which quantifies the cosine of the angle between these vectors in high-dimensional space. Specifically, it takes normalized feature embedding where each vector's magnitude scaled to one and calculates the dot product between the query vector and each candidate image vector. The resulting scalar value ranges from -1 to 1, indicating the degree of similarity: values closer to 1 denote high relevance, while those near -1 suggest dissimilarity. By performing this calculation for all candidate images, the unit ranks them in descending order based on their cosine similarity scores, effectively prioritizing images most similar to the query in terms of their feature representations. This ranking facilitates efficient retrieval or matching tasks by identifying the images that best correspond to the query's content.

[0027] To the similarity measurement unit, a retrieval and ranking unit is linked that retrieves and ranks images based on similarity scores. The retrieval and ranking unit functions by first collecting the similarity scores computed between the query image and a set of candidate images, obtained through the cosine similarity. It then sorts these candidate images in descending order of their similarity scores, so that the most relevant images those with the highest similarity to the query appear at the top of the list. This process often involves efficient sorting protocols that handle large datasets quickly. The unit also incorporate threshold to filter out images below a certain relevance level, or apply additional ranking refinements using learned weights or re-ranking strategies to improve retrieval accuracy.

[0028] After retrieving and ranking the images based on similarity scores, a display Unit linked to the Retrieval and Ranking Unit displays ranked images to the user. The display unit provides a visualization dashboard displaying precision, recall, and F1-score metrics for retrieval performance. The display unit functions as the visualization interface that presents the ranked images retrieved by the retrieval and ranking unit alongside performance metrics such as precision, recall, and F1-score, which are calculated based on the comparison between the retrieved set and ground truth relevant images. It receives the ranked list of images and the associated metrics from the system, then organizes and visualizes this data in an intuitive dashboard format showing thumbnails of top-ranked images, bar charts or line graphs for precision and recall at various retrieval thresholds, and composite metrics like F1-score to summarize overall performance.

[0029] The present invention works best in the following manner, the user-interface installed in the computing unit wirelessly linked with the system via the communication module, accessed by the user to uploaded input data, that is transmitted to the pre-Processing Unit configured to input images, normalizes and augments images to prepare for feature extraction. The U Net Feature Extraction has an encoder-decoder architecture that includes skip connections to preserve spatial context during feature extraction. The re- weighting unit dynamically adjusts feature weights based on query relevance by using a query-aware protocol to prioritize texture or shape features based on the query image content. The similarity measurement unit calculates cosine similarity to rank images in descending order of relevance to the query. The retrieval and ranking unit retrieves and ranks images based on similarity scores. The display Unit linked to the Retrieval and Ranking Unit displays ranked images to the user. The display unit provides a visualization dashboard displaying precision, recall, and F1-score metrics for retrieval performance.

[0030] 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 system for retrieving images based on visual content, comprising:

i) a user-interface installed in a computing unit wirelessly linked with the system via a communication module, that is accessed by a user to uploaded input data, that is transmitted to a pre-Processing Unit configured to input images, normalizes and augments images to prepare for feature extraction;
ii) a U Net Feature Extraction Unit linked with Pre-Processing Unit, to extract low-level and high-level features using a U Net architecture;
iii) a feature re-weighting unit inked to the U Net Feature Extraction Unit, dynamically adjusts feature weights based on query relevance;
iv) a similarity measurement unit connected to the feature re-weighting Unit, calculates cosine similarity to rank images;
v) a retrieval and ranking unit linked to the similarity measurement unit, retrieves and ranks images based on similarity scores; and
vi) an display Unit linked to the Retrieval and Ranking Unit, displays ranked images to the user.

2) The device as claimed in claim 1, wherein U Net Feature Extraction has an encoder-decoder architecture that includes skip connections to preserve spatial context during feature extraction.

3) The device as claimed in claim 1, wherein the feature re-weighting unit uses query-aware arrangements to prioritize texture or shape features based on the query image content.

4) The device as claimed in claim 1, wherein the similarity measurement unit cosine similarity scores are used to rank images in descending order of relevance to the query.

5) The device as claimed in claim 1, wherein system further includes a user feedback arrangements connected to the feature re-weighting unit to refine feature weights based on user input.

6) The device as claimed in claim 1, wherein the display unit provides a visualization dashboard displaying precision, recall, and F1-score metrics for retrieval performance.