Image search system and image search method

Abstract

To provide an image search system or the like that determines an image from a language expression of a visual feature for the image and appropriately generates a response content. The image search system includes a preprocessing unit, an acquisition unit, a specification unit, and an output unit. The preprocessing unit generates at least one image description text representing a visual feature for each of a plurality of images, and stores the image description text in association with each of the plurality of images. The acquisition unit acquires input information including a language expression of a visual feature for an image. The specification unit matches a corresponding image description text based on the input information, and determines an image associated with the image description text. The output unit generates a response content for the image and outputs the response content. In addition, an artificial intelligence using a cross-modal technology can be used when generating the image description text.

Description

Technical Field

[0001] This invention relates to an image search system and an image search method. Background Technology

[0002] Conventional image search systems require users to directly input the referenced image when searching for images, icons, or graphics. However, it is difficult to input the image into the system when the referenced image is a printed document or a display on another terminal that cannot be captured. Patent Document 1 discloses a similar image search device to prevent a decrease in search accuracy.

[0003] Patent Document 1: Japanese Patent Application Publication No. 2011-203776 Summary of the Invention

[0004] The similar image search device described in Patent Document 1 uses not only the similarity of image feature quantities with a specified reference image, but also the similarity based on text feature quantities related to text associated with the specified reference image. This prevents a decrease in search accuracy when searching for similar images using an image as the search key. However, the similar image search device described in Patent Document 1 is a device for searching for similar images from a reference image, and does not describe searching for image content based on the user's linguistic description of the image. Therefore, it is impossible to generate an answer based on a question containing a linguistic description of the image.

[0005] In view of the above-mentioned issues, the present invention provides an image search system that determines an image based on a linguistic description of the visual features of the image and appropriately generates response content.

[0006] The image search system according to one aspect of the present invention includes a preprocessing unit, an acquisition unit, a selection unit, and an output unit. The preprocessing unit generates at least one image description representing visual features for each of multiple images, and associates and stores the image descriptions with each of the multiple images. The acquisition unit acquires input information containing linguistic descriptions of the visual features of the images. The selection unit matches the corresponding image descriptions based on the input information and determines the images associated with the image descriptions. The output unit generates and outputs response content for each image.

[0007] In the aforementioned image search system, the preprocessing unit can extract image attributes related to visual features from each of multiple images, generate image descriptions containing these attributes, associate the image attributes with each of the multiple images, and then store them. The image search system may also include an attribute extraction unit, which extracts search attributes related to visual features from the input information, and a further matching unit matches the corresponding image descriptions based on these search attributes.

[0008] In the aforementioned image search system, the image can be an icon, symbol, sign, or marker. The input information includes a question related to the icon, symbol, sign, or marker. The attribute extraction unit extracts search attributes from the question. In the image search system, the identification unit can determine image attributes that match the search attributes and identify images associated with image descriptions containing the determined image attributes. The output unit generates an answer to the question as the response content.

[0009] In the aforementioned image search system, the preprocessing unit can generate image generation prompts based on images possessing image attributes. In the image search system, the selection unit can generate a search image based on prompts corresponding to search attributes, and determine the corresponding image based on the search image.

[0010] In one aspect of the image search method of the present invention, the following processes are performed by a computer: The computer generates at least one image description representing visual features for each of multiple images, and associates and stores the image descriptions with each of the multiple images. The computer acquires input information containing a linguistic description of the visual features of the image. The computer matches the corresponding image description based on the input information and determines the image associated with the image description. The computer generates and outputs a response for the image.

[0011] Invention Effects

[0012] According to the present invention, an image search system and image search method can be provided to determine an image based on a linguistic description of the visual features of the image and to appropriately generate a response. Attached Figure Description

[0013] Figure 1 This is a block diagram of the image search system involved in Implementation Method 1.

[0014] Figure 2 This is a flowchart of the image search method involved in Implementation Method 1.

[0015] Figure 3 This is a block diagram of the image search system involved in Implementation Method 2.

[0016] Figure 4 This is a flowchart of the image search method involved in Implementation Method 2.

[0017] Figure 5 This is a flowchart of the image search method involved in Implementation Method 3.

[0018] Figure 6 It is a block diagram illustrating the hardware structure of a computer. Detailed Implementation

[0019] The present invention will now be described through embodiments thereof, but the invention as described in the claims is not limited to these embodiments. Furthermore, not all structures described in the embodiments are necessarily necessary as means of solving the problem. For clarity, appropriate omissions and simplifications have been made in the following descriptions and drawings. Additionally, in the drawings, the same symbols are used to denote the same elements, and repeated descriptions have been omitted as necessary.

[0020] <Implementation Method 1>

[0021] refer to Figure 1 The image search system 10 according to Embodiment 1 will be described. Figure 1 This is a block diagram of the image search system 10 according to Embodiment 1. The image search system 10 determines an image based on a linguistic description of its visual features and generates a response for the determined image. Here, the image may be, for example, an icon, symbol, mark, or tag. The image is not limited to these and may also be a photograph. Furthermore, the visual features are the visual characteristics possessed by the image, such as the image's shape, size, color, or orientation. The image search system 10 includes a preprocessing unit 101, an acquisition unit 102, a selection unit 103, and an output unit 104.

[0022] The preprocessing unit 101 generates at least one image description representing the visual features of each of the multiple images. The preprocessing unit 101 may also divide the visual features into multiple image descriptions. The image description is a text explaining the image. Image descriptions are also called captions. Image descriptions are generated, for example, using generative artificial intelligence (AI). Here, generative AI is a multimodal model based on large-scale language generation technology.

[0023] Multimodal models, like Large Language Models (LLMs), fully utilize deep learning and have learned not only for text processing but also for images, speech, or video. In this context, a multimodal model has at least been learned for both images and text, enabling it to convert between the two.

[0024] The preprocessing unit 101 includes a storage unit (not shown) that associates image descriptions with each of multiple images and stores them thereafter. For example, the preprocessing unit 101 stores a list of images associated with their descriptions as a database (DB). Furthermore, the preprocessing unit 101 may not store the images themselves, but instead store the image's source or identification information. Additionally, the preprocessing unit 101 may also associate the meaning represented by the image with the image and store it thereafter.

[0025] The acquisition unit 102 acquires input information containing a linguistic description of the visual features of the image. The acquisition unit 102 acquires the input information in text format. The acquisition unit 102 receives user questions, for example, via an interface such as a communication device, microphone, or touch panel, and acquires them as text-formatted input information. Here, the question might be, for example, asking the user about the meaning of the image they are referring to.

[0026] The specific unit 103 matches the corresponding image description text based on the input information and determines the image associated with the image description text. For example, the specific unit 103 compares the linguistic descriptions and image description texts containing visual features in the input information, and matches the image description texts corresponding to the linguistic descriptions based on similarity. The specific unit 103 may extract specific words contained in the linguistic descriptions of visual features and match the corresponding image description texts based on these specific words. The specific unit 103 may use algorithms or AI, for example, to compare the linguistic descriptions and image description texts.

[0027] The output unit 104 generates and outputs response content for the determined image. The output unit 104 may include, for example, image data, the image's source, an image description, or the generated response in the response content. That is, the response content can be data from the determined image or an image-based response. Furthermore, the output unit 104 can calculate the degree of consistency between the input information and the image description and include it as a confidence level in the response content.

[0028] Therefore, the image search system 10 can obtain input information from the user in text format and determine the image corresponding to the visual features represented by the input information. Thus, the image search system 10 can determine the image requested by the user and generate a response based on the determined image.

[0029] Figure 2 This is a flowchart of the image search method according to Embodiment 1. The flowchart of the image search method according to the image search system 10 includes steps S11 to S16.

[0030] In step S11, the preprocessing unit 101 of the image search system 10 generates at least one image description representing visual features for each of the multiple images. Here, the image may be, for example, a mark, icon, symbol, or sign. Furthermore, the visual features may be the shape, size, color, or characters or symbols contained in the image. In step S12, the preprocessing unit 101 associates the image description with each of the multiple images and stores it. Additionally, preprocessing is performed in steps S11 and S12. Therefore, when the image search system 10 executes the image search method according to Embodiment 1 multiple times, the processing in steps S11 and S12 can be omitted in subsequent iterations.

[0031] In step S13, the acquisition unit 102 acquires input information in text format, which includes a linguistic description of the visual features of the image referenced by the user. Here, the input information may be, for example, a question about the meaning represented by the image. The acquisition unit 102 receives, for example, communication, voice, or typing input from the user and acquires it as text-formatted input information.

[0032] In step S14, the targeting unit 103 matches the corresponding image description text based on the input information. For example, the targeting unit 103 compares the linguistic expression of the visual features represented by the input information with multiple stored image description texts, and matches the image description text corresponding to the linguistic expression of the visual features based on similarity. The matching method of the targeting unit 103 is not limited to this. In step S15, the targeting unit 103 determines the image associated with the image description text matched in step S14.

[0033] In step S16, the output unit 104 generates and outputs a response to the determined image. The response may be, for example, an answer to a question that is input information. The output unit 104 may output the determined image. The output unit 104 may output the response using a visual or auditory prompting unit. The output unit 104 is not limited to this; it may also output the response by sending it to an external device.

[0034] As explained above, the image search system 10 determines the image associated with the image description by matching the linguistic description of the visual features contained in the acquired input information. Thus, the image search system 10 can identify the image corresponding to the input information and generate appropriate response content for the image. Therefore, the user of the image search system 10 can obtain image-related search results without inputting an image into the image search system 10.

[0035] Alternatively, the image search system 10 may have a processor and a storage device (not shown). The storage device included in the image search system 10 may include, for example, a storage device containing non-volatile memory such as flash memory or a solid-state drive (SSD). In this case, the storage device stores a computer program (hereinafter also simply referred to as the program) for performing the above-described method. The processor then reads the computer program from the storage device into a buffer memory such as dynamic random access memory (DRAM) and executes the program.

[0036] Each structure of the image search system 10 can be implemented by dedicated hardware. Furthermore, some or all of each component can be implemented using general-purpose or dedicated circuitry, processors, or combinations thereof. They can be composed of a single chip or multiple chips connected via a bus. Some or all of each component of each device can be implemented using the aforementioned circuitry and programs. The processor can be a central processing unit (CPU), a graphics processing unit (GPU), a field-programmable gate array (FPGA), or the like. Moreover, at least a portion of the processing performed by the image search system 10 can be provided as Software as a Service (SaaS). Additionally, the description of the structures described herein is also applicable to other systems described below in this invention.

[0037] Furthermore, the image search system 10 can be implemented by hardware mounted on the mobile body, or some of the hardware can be located outside the mobile body. In this case, the device located outside the mobile body (e.g., a server) and the device located on the mobile body are connected via a communication network.

[0038] <Implementation Method 2>

[0039] Figure 3 This is a block diagram of the image search system 11 according to Embodiment 2. The image search system 11 extracts search attributes related to visual features from the questions contained in the input information, and further matches image descriptions based on the search attributes. Here, the questions are related to icons, signs, symbols, or markings. The image search system 11 according to Embodiment 2 and... Figure 1 The image search system 10 has a partially identical structure. Therefore, the description of the structure of the image search system 11, which performs the same processing as the image search system 10, is omitted. The image search system 11 includes a preprocessing unit 111, an acquisition unit 102, an attribute extraction unit 113, a specification unit 114, and an output unit 115.

[0040] The preprocessing unit 111 extracts image attributes related to visual features from each of the multiple images. Here, the images are icons, logos, symbols, or markers. Image attributes are identifiers that represent the visual features of the image. Image attributes may include, for example, the image's color, background color, or orientation, or the number of symbols, characters, or constituent elements contained in the image.

[0041] The preprocessing unit 111 generates image descriptions containing image attributes. Specifically, the preprocessing unit 111 generates appropriate image descriptions by constructing prompts containing image attributes and inputting them into a multimodal model. The preprocessing unit 111 performs, for example, prompt fine-tuning. Prompt fine-tuning is a method of optimizing prompts for obtaining the desired output through learning, without requiring relearning or adjustment of the generative AI. Furthermore, the prompt is an instruction that prompts the generative AI to output the desired result. Here, the preprocessing unit 111 can generate one image description representing multiple image attributes, or it can generate multiple image descriptions representing each of the multiple image attributes individually.

[0042] The preprocessing unit 111 associates image attributes with each of the multiple images and stores them. For example, the preprocessing unit 111 may further associate image attributes with a list that associates images and image descriptions and store this list as a database (DB). Furthermore, the preprocessing unit 111 may additionally generate and store a list that associates images with image attributes.

[0043] The attribute extraction unit 113 extracts search attributes related to visual features from the input information. Search attributes correspond to image attributes and are identifiers representing the visual features of the image referenced by the user. Here, the attribute extraction unit 113 extracts search attributes from the question contained in the input information. That is, the attribute extraction unit 113 extracts search attributes from the linguistic description of the visual features of the image in the question.

[0044] The specific unit 114 further matches the corresponding image description text based on the search attributes extracted by the attribute extraction unit 113. Specifically, the specific unit 114 determines candidates for corresponding image description text based on the search attributes, and matches the corresponding image description text from the candidates based on the input information. That is, the specific unit 114 determines image attributes that match the search attributes extracted from the question, and determines images that are associated with image description texts containing the determined image attributes. Thus, the image search system 11 can appropriately determine the image description text and image corresponding to the input information by pre-defining candidates based on search attributes.

[0045] The output unit 115 generates and outputs an answer to the question as response content. The output unit 115 may include a prompting unit (not shown) such as a display or a speaker, and prompt the user with the response content visually or audibly. The output unit 115 is not limited to this, and may also output the response content by sending it to an external device.

[0046] Specifically, the image search system 11 installed in the vehicle receives questions related to interior components or instruments of the car. These questions may include inquiries about the names or meanings of warning lights or icons, devices or components, or buttons on the interior instruments. Furthermore, questions may include inquiries about the names or meanings of traffic signs, traffic signals, or shop signs. Additionally, questions may include inquiries about the meanings of arrows, abbreviations, or Point of Interest (POI) icons in in-vehicle navigation or map applications.

[0047] Figure 4 This is a flowchart of the image search method according to Embodiment 2. The flowchart of the image search method according to the image search system 11 includes steps S21 to S28.

[0048] In step S21, the preprocessing unit 111 extracts image attributes related to visual features from each of the multiple images. Here, the images are icons, logos, symbols, or markers. Image attributes include, for example, the color of the image, its background color or orientation, or the number of symbols, characters, or constituent elements contained in the image. In step S22, the preprocessing unit 111 generates image descriptions containing the image attributes for each of the multiple images. The preprocessing unit 111, for example, performs fine-tuning to ensure that the image descriptions include the content of the image attributes. In step S23, the preprocessing unit 111 associates the image attributes and image descriptions with each of the multiple images and stores them.

[0049] Furthermore, preprocessing is performed in steps S21 to S23. Therefore, when the image search system 11 executes the image search method according to Embodiment 2 multiple times, the processing in steps S21 to S23 after the second execution can be omitted.

[0050] In step S24, the acquisition unit 102 acquires input information containing a linguistic description of the visual features of the image. Here, the input information includes questions related to icons, signs, symbols, or markers. Furthermore, the input information is acquired in text format. The acquisition unit 102 receives, for example, communication, voice, or typing input from a user via an interface and acquires it as input information. Next, in step S25, the attribute extraction unit 113 extracts search attributes related to the visual features from the input information. The search attributes correspond to the image attributes extracted in step S21.

[0051] In step S26, the targeting unit 114 matches the corresponding image description text based on the search attributes and input information. Then, in step S27, the targeting unit 114 determines the image associated with the image description text matched in step S26. That is, the targeting unit 114 determines the image attributes that match the search attributes and determines the image associated with the image description text containing the determined image attributes.

[0052] The specificity unit 114 filters the image description text, for example, using image attributes that match the search attributes. Next, the specificity unit 114 matches the questions and image description text contained in the input information based on the arrangement or connection relationships of words representing the search attributes. Thus, the specificity unit 114 can narrow down the corresponding image description text according to the search attributes. The specificity unit 114 can use AI such as LLM for matching.

[0053] In step S28, the output unit 115 generates and outputs response content for the determined image. Here, the response content is an answer to the question contained in the input information. The output unit 115 outputs the response content by prompting the user visually or audibly. The output unit 115 is not limited to this; it may also output the response content by sending it to an external device.

[0054] Image search system 11 extracts the image attributes "black," "yellow," "A," and "OFF" from an image of an indicator light related to the idle stop function, specifically from the image of a "yellow-lit mark on a black background with OFF written below an arrow surrounding an A." Next, image search system 11 generates a description for the indicator light image: "A symbol containing the characters 'A' and 'OFF' emphasized in yellow and black." If a user inputs "What does the yellow-lit 'A' mark mean?" into image search system 11, image search system 11 extracts the search attributes "yellow" and "A" from the question. Then, based on the search attributes and the question, image search system 11 determines the image of the indicator light, generates an answer such as "Indicates an abnormality in the idle stop system," and outputs it.

[0055] As explained above, the image search system 11 reduces the processing associated with image identification by narrowing down the candidate image descriptions based on search attributes. Therefore, the image search system 11 can identify images based on linguistic descriptions of their visual features and generate more appropriate answers to questions. Consequently, the image search system 11 can appropriately answer the questions entered by the user.

[0056] Furthermore, since the generative AI has learned in a language different from the input information, the image search system 11 can translate the explanatory text and input information into the primary learning language of the generative AI. Thus, the image search system 11 can maximize the performance of the generative AI it utilizes, appropriately identify the image, and generate a response.

[0057] Furthermore, when matching image descriptions based on input information, the image search system 11 can pre-compare multiple matching methods and determine the best method or methods that are among the multiple candidates. Thus, the image search system 11 is able to more appropriately identify images and generate responses.

[0058] <Implementation Method 3>

[0059] Figure 5 This is a flowchart of the image search method according to Embodiment 3. The image search method according to Embodiment 3 extracts search attributes related to visual features from input information, generates an image for searching (i.e., a search image) based on the search attributes, and matches the stored image with the search image. The image search system executing the image search method according to Embodiment 3 has... Figure 3 The image search system 11 in Embodiment 3 has the same structure. Therefore, the description of the structure of the image search system that performs the same processing as the image search system 11 is omitted. The image search method of the image search system according to Embodiment 3 includes steps S31 to S37.

[0060] In step S31, the preprocessing unit extracts image attributes related to visual features from each of the multiple images. In step S32, the preprocessing unit generates image generation prompts based on the images with image attributes. Furthermore, preprocessing is performed in steps S31 and S32. Therefore, when the image search system executes the image search method according to Embodiment 3 multiple times, the processing in steps S31 and S32 can be omitted in subsequent iterations.

[0061] In step S33, the acquisition unit acquires input information containing a linguistic description of the visual features of the image. Here, the input information includes questions related to icons, signs, symbols, or markers. Furthermore, the input information is acquired in text format. Next, in step S34, the attribute extraction unit extracts search attributes related to the visual features from the input information. The search attributes correspond to the image attributes extracted in step S31.

[0062] In step S35, the targeting unit generates a search image based on image generation prompts corresponding to the search attributes. Next, in step S36, the targeting unit determines the corresponding image based on the search image. That is, the targeting unit determines image attributes that match the search attributes and determines the image corresponding to the input information by matching it with the search image generated based on the prompts corresponding to the determined image attributes. Thus, the targeting unit can apply image matching techniques to improve image determination accuracy.

[0063] Furthermore, in step S35, the targeting unit can make the search image closer to the user's recognition through dialogue with the user. The targeting unit may, for example, use a display unit such as a monitor to display the generated search image and request the user's confirmation. At this time, the targeting unit acquires correction information indicating the correction area pointed out by the user via the acquisition unit. The targeting unit may, for example, acquire the correction information by detecting at least one of voice input and touch input via the acquisition unit. Next, the targeting unit may, for example, correct the original input information to generate a search image consistent with the correction information. Alternatively, the targeting unit may generate additional prompts for correcting the search image based on the correction information and generate a corrected search image. Thus, the targeting unit is able to generate a search image that closely approximates the user's recognition.

[0064] Furthermore, the targeting unit can generate multiple search images in step S35 and present them to the user using a display unit such as a monitor. At this time, the user selects the search image closest to the identified image from the multiple search images, and the targeting unit receives the user's selection via the acquisition unit. Thus, the targeting unit can appropriately generate a search image that closely approximates the user's identified image.

[0065] In step S37, the output unit generates and outputs response content for the determined image. Here, the response content is the answer to the question contained in the input information.

[0066] As explained above, the image search system according to Embodiment 3 extracts search attributes related to visual features from input information containing a question, and generates a search image based on the search attributes. Therefore, the image search system can use image matching technology to determine the image corresponding to the input information. Consequently, the image search system can appropriately generate a response for the image corresponding to the input information.

[0067] Furthermore, the present invention is not limited to the above embodiments and can be appropriately modified without departing from the spirit of the invention. For example, the image search method according to Embodiment 2 and the image search method according to Embodiment 3 can be executed simultaneously to combine the answer content. Moreover, the image search system can include a user authentication unit that authenticates users through voice, biometrics, or personal authentication, and stores each user's descriptive characteristics as characteristic information. The user authentication unit can authenticate users by allowing them to select a registered account. Here, descriptive characteristics include, for example, the user's habits or preferences regarding the way they describe shape, color, or size, and word order.

[0068] Here, the acquisition and specificity departments obtain characteristic information associated with authenticated users from the user authentication department, and convert user input into a standard expression easily processed by the image search system based on this characteristic information. Specifically, the acquisition and specificity departments convert expressions such as "gray," "mouse color," or "lead color," which are different for each user, into the standard expression "gray" based on the characteristic information. Thus, the image search system can convert different expression methods for each user into a unified expression for processing. Therefore, the image search system can appropriately generate and output response content based on the input information, regardless of the user's expression characteristics.

[0069] Additionally, the output unit can obtain characteristic information associated with authenticated users from the user authentication unit. For example, based on this characteristic information, the output unit converts the standard expression used during processing into a expression tailored to each user's specific characteristics before outputting it. Thus, the image search system can output responses in a way that is familiar to the user.

[0070] <Examples of hardware architecture>

[0071] The following describes examples of how the functional structures of the image search system in this invention are implemented using a combination of hardware and software.

[0072] Figure 6 This is a block diagram illustrating the hardware structure of a computer. The image search system of this invention can achieve the above-mentioned functions using a computer 500 including the hardware structure shown in the figure. The computer 500 can be a portable computer such as a smartphone or tablet terminal, or a stationary computer such as a PC. The computer 500 can be a dedicated computer designed to implement various devices, or a general-purpose computer. By installing the prescribed application programs, the computer 500 can achieve the desired functions.

[0073] Computer 500 includes a bus 502, a processor 504, a memory 506, a storage device 508, an input / output interface (I / F) 510, and a network interface (I / F) 512. Bus 502 is a data transmission path for the processor 504, memory 506, storage device 508, input / output interface 510, and network interface 512 to send and receive data. The method of connecting the processor 504 and other components to each other is not limited to bus connection.

[0074] Processor 504 is a processor such as a CPU, GPU, or FPGA. Memory 506 is a main storage device implemented using random access memory (RAM) or similar methods.

[0075] Storage device 508 is an auxiliary storage device implemented using a hard disk, SSD, memory card, or read-only memory (ROM). Storage device 508 stores programs for implementing desired functions. Processor 504 implements the various functional units of each device by reading these programs into memory 506 and executing them.

[0076] Input / output interface 510 is an interface for connecting computer 500 and input / output devices. For example, input devices such as keyboards or output devices such as displays can be connected to input / output interface 510. Network interface 512 is an interface for connecting computer 500 to a network.

[0077] Symbol Explanation

[0078] 10, 11 - Image search system; 101, 111 - Preprocessing unit; 102 - Acquisition unit; 103, 114 - Specific unit; 104, 115 - Output unit; 113 - Attribute extraction unit; 500 - Computer; 502 - Bus; 504 - Processor; 506 - Memory; 508 - Storage device; 510 - Input / output interface; 512 - Network interface.

Claims

1. An image search system characterized by comprising: It has a preprocessing unit, an acquisition unit, a specific unit, and an output unit. The preprocessing unit performs the following processing: Generate at least one image description representing the visual features of each of the multiple images; and The image description text is associated with each of the multiple images and then stored. The acquisition unit acquires input information containing a linguistic representation of the visual features of the image. The specific unit matches the corresponding image description text based on the input information, and determines the image associated with the image description text. The output unit generates and outputs a response to the image.

2. The image search system according to claim 1, characterized in that, The preprocessing unit performs the following processing: Extract image attributes related to the visual features from each of the multiple images; The generated content includes the image description text containing the image attributes; and The image attributes are associated with each of the multiple images and then stored. The image search system also includes an attribute extraction unit, which extracts search attributes related to the visual features from the input information. The specific part further matches the corresponding image description text based on the search attribute.

3. The image search system according to claim 2, characterized in that, The image is an icon, symbol, sign, or marker. The input information includes questions related to the icon, symbol, sign, or mark. The attribute extraction unit extracts the search attribute from the question. The specific unit determines the image attribute that matches the search attribute, and determines the image associated with the image description containing the determined image attribute. The output unit generates an answer to the question as the answer content.

4. The image search system according to claim 2, characterized in that, The preprocessing unit generates image generation prompts in the image attributes based on the image having the image attributes. The specific part is processed as follows: A search image is generated based on the prompt corresponding to the search attribute; and The corresponding image is determined based on the search image.

5. An image search method characterized by, The computer performs the following processing: Generate at least one image description representing the visual features of each of the multiple images; The image description text is associated with each of the multiple images and then stored. Obtain input information containing a linguistic representation of the visual features of the image; Based on the input information, match the corresponding image description text, and determine the image associated with the image description text; and Generate and output the response content based on the image.

Images