Video character relationship construction method and device, and medium

Abstract

The embodiment of the invention provides a video character relationship construction method, which is applied to the technical field of video processing and analysis, and comprises the following steps: performing data processing on a target video to obtain video structure information of the target video, and dividing the target video into at least two content levels according to the video structure information, and executing step-by-step abstract text generation from the lowest content level to the higher content level. And then, inputting the obtained highest content hierarchy abstract text containing the complete narrative logic into a pre-trained abstract generation model, and converting the structured information after hierarchy fusion into a coherent and refined plot abstract text. And finally, inputting the plot summary text into the same summary generation model again, and utilizing a target instruction for indicating generation of character relationship information, so that the model can understand and reasone the character relationship based on the deep semantics of the whole plot and output the character relationship, thereby realizing the precise construction of the dynamic character relationship.

Description

Technical Field

[0001] This disclosure relates to the field of video processing and analysis technology, and in particular to a method, device and medium for constructing relationships between people in a video. Background Technology

[0002] With the widespread adoption of long-form video content such as TV dramas and series on streaming platforms, users face the challenge of quickly and accurately understanding complex character relationships and their evolution when confronted with dozens or even hundreds of hours of storyline. Whether it's ordinary viewers making viewing decisions or professionals engaging in content creation or marketing analysis, there is an urgent need for automated, in-depth analysis of the dynamic relationships between characters in videos. Therefore, how to automatically construct character relationships with temporal evolution characteristics from massive amounts of video data has become a key technical challenge for enhancing video content understanding and intelligent application value.

[0003] Currently, existing technologies for understanding relationships between characters in videos primarily employ separate processing methods. On one hand, computer vision-based techniques, such as face detection and clustering, can identify and track different characters appearing in a video, but can only generate a list of characters appearing, failing to reveal the social relationships between them. On the other hand, natural language processing-based techniques, such as extracting co-occurrence relationships from subtitles or dialogue text, can analyze shallow connections between characters. These techniques typically process visual or textual cues in isolation, generating static and fragmented results.

[0004] Existing technologies cannot effectively integrate multimodal information and perform deep semantic reasoning, resulting in serious deficiencies in the core task of character relationship generation, which in turn leads to a poor user experience. Summary of the Invention

[0005] To solve the above-mentioned technical problems, or at least partially solve them, this disclosure provides a method, device, and medium for constructing character relationships in a video.

[0006] This disclosure provides a method for constructing character relationships in a video. The method includes: extracting data from the target video to obtain a list of shot transition timestamps, subtitle information, character information, and a list of scene transition timestamps, providing detailed underlying data support for subsequent analysis. Then, based on the shot transition timestamps, the target video can be broken down into multiple shot segments, and precise shot-level summary texts are generated by combining the subtitle and character information within each shot segment, achieving a preliminary understanding and summary of the content of each shot. Subsequently, consecutive shot segments can be aggregated into scenes using scene transition timestamps, and the summary texts and subtitle information of all shots within a scene can be merged to generate scene-level summary texts, further refining the core plot and emotional trajectory of the scene. Finally, all scene summary texts are aggregated and input into a deeply trained summary generation model, comprehensively considering the logical connections and emotional context between scenes to generate a complete plot summary text covering the key plot, character development, and emotional conflicts of the target video. This process, through multi-level and multi-dimensional data integration and analysis, effectively compensates for the shortcomings of existing technologies, achieves deep semantic understanding of complex videos, accurately presents core characters and their dynamic relationships, and meets users' dual needs of quickly grasping the overall plot and deeply analyzing plot details.

[0007] This disclosure also provides a computing device, the computing device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the video character relationship construction method provided in this disclosure.

[0008] This disclosure also provides a computer-readable storage medium storing a computer program for executing a video character relationship construction method as provided in this disclosure. Attached Figure Description

[0009] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0010] Figure 1 A flowchart illustrating a method for constructing character relationships in a video, as provided in an embodiment of this disclosure; Figure 2 A schematic diagram of the structure of a video character relationship construction device provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of the present disclosure. Detailed Implementation

[0011] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0012] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0013] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0014] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0015] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0016] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0017] With the widespread adoption of long-form video content such as TV dramas and series on streaming platforms, users face the challenge of quickly and accurately understanding complex character relationships and their evolution when confronted with dozens or even hundreds of hours of storyline. Whether it's ordinary viewers making viewing decisions or professionals engaging in content creation or marketing analysis, there is an urgent need for automated, in-depth analysis of the dynamic relationships between characters in videos. Therefore, how to automatically construct character relationships with temporal evolution characteristics from massive amounts of video data has become a key technical challenge for enhancing video content understanding and intelligent application value.

[0018] Currently, existing technologies for understanding relationships between characters in videos primarily employ separate processing methods. On one hand, computer vision-based techniques, such as face detection and clustering, can identify and track different characters appearing in a video, but can only generate a list of characters appearing, failing to reveal the social relationships between them. On the other hand, natural language processing-based techniques, such as extracting co-occurrence relationships from subtitles or dialogue text, can analyze shallow connections between characters, but struggle to infer complex relationship types and their dynamic changes. These techniques typically process visual or textual cues in isolation, generating static and fragmented results.

[0019] The fundamental problem with the aforementioned existing technologies is that they cannot achieve effective fusion of multimodal information and deep semantic reasoning, resulting in serious deficiencies in the core task of generating character relationships.

[0020] In view of this, this application proposes a method for constructing character relationships in videos. This application processes the target video to obtain its video structure information (such as shots, scene boundaries, subtitles, and character identifiers), establishing a unified multimodal data foundation for subsequent analysis and ensuring the comprehensiveness of information sources. Subsequently, based on this video structure information, the target video is divided into at least two content levels (shots, scenes, etc.) (each level includes multiple video segments), and hierarchical summary text generation is performed from the lowest content level (shots) to higher content levels (scenes, episodes, etc.). In this process, the summary text of each higher level is generated by integrating the semantics of its subordinate lower-level summary texts. This fundamentally achieves deep semantic integration of multimodal information such as visual content, character dynamics, and dialogue text, effectively overcoming the problem of information silos. Then, the resulting highest-level summary text containing the complete narrative logic (i.e., the plot summary text) is input into a pre-trained summary generation model. This model, leveraging its powerful natural language understanding and generation capabilities, transforms the hierarchically fused structured information into a coherent and concise plot summary text, achieving a qualitative leap from fragmented facts to a holistic narrative. Finally, this plot summary text is input into the same summary generation model again, and using the target instruction to generate character relationship information, the model, based on a deep semantic understanding of the overall plot, infers and outputs character relationship information that includes not only the type of relationship between characters but, more importantly, the temporal changes in that relationship type as the plot progresses. This achieves precise construction of dynamic character relationships.

[0021] To address the aforementioned issues, this disclosure provides a method for constructing character relationships in videos, which will be described below with reference to specific embodiments.

[0022] Figure 1This is a flowchart illustrating a method for constructing character relationships in a video according to an embodiment of this disclosure. This method can be executed by a video character relationship construction device, which can be implemented using software and / or hardware, and is generally integrated into a computing device. Figure 1 As shown, the method includes: S101: Perform data processing on the target video to obtain the video structure information of the target video.

[0023] The computing device can extract data from the target video to obtain the video structure information of the target video.

[0024] For example, data extraction from a target video by a computing device may include: shot recognition, i.e., identifying the time point of each shot transition in the target video and obtaining a list of timestamps for shot transitions; subtitle extraction, i.e., obtaining subtitle information for the target video; character recognition, i.e., obtaining character information for the target video; and scene detection, i.e., identifying the time point of each scene transition in the target video and obtaining a time sequence table of scene transitions for the target video. In other words, the video structure information may include a list of timestamps for shot transitions, subtitle information, character information, and a time sequence table of scene transitions.

[0025] It's important to explain that the timestamp list of shot transitions records the specific time points of each shot transition in the target video. This can be used to analyze the rhythm and narrative structure of the shots.

[0026] The subtitle information includes all subtitle texts appearing in the target video and their display time range, which can be used to understand the dialogue content and the time correspondence.

[0027] Character information refers to the process of binding facial clustering to the names of people appearing in a target video, which can be used to identify and track different characters in the video.

[0028] The timestamp list for scene transitions is used to mark the time points of scene transitions in the target video, in order to distinguish different story scenes or plot units.

[0029] For example, if the input target video is an episode of a TV series, the computing device can extract data from the video data of that episode.

[0030] For the video data of this set, shot recognition can be, for example, reading the video data at a fixed frame rate or frame by frame, and extracting the picture information of each frame. Then, similarity analysis can be performed on adjacent frames. The similarity between them can be evaluated by calculating the pixel difference (such as mean square error) or feature vector difference (such as features extracted using a convolutional neural network) between the frames. When the similarity is lower than the set threshold, it can be determined as a shot transition point, and the timestamp of this frame is recorded. Repeat this process until the entire video data is processed, and finally generate a list of timestamps including all shot transition time points, thus completing the recognition of shot transitions and obtaining the list of timestamps of shot transitions in this video data.

[0031] For another example, in the embodiments of this application, the video data can also be input into the TransNetV2 model. This model can automatically analyze the video frame by frame and identify the key frames of shot transitions. TransNetV2 detects the significant changes between frames through deep learning technology to determine the position of shot transitions. The result output by this model can be a probability distribution, indicating the possibility that each frame is a shot transition point. According to the set threshold (such as a probability greater than 0.5), the frames with high probability are selected as shot transition points, and the timestamps of these frames are obtained. Finally, these timestamps are sorted into a list, such as shot_timestamps = [0.0, 15.2, 25.8,...], where each timestamp represents the time point of a shot transition, thus completing the generation of the list of shot transition timestamps.

[0032] For subtitle extraction from the video data of this set, it can be, for example, decomposing the video data frame by frame to obtain a series of static image frames. Then, PaddleOCR can be used to perform optical character recognition on each frame image to scan and extract the text content in the picture. During the extraction process, the timestamp information of each frame can be recorded to determine the start and end times of the subtitles. By analyzing the appearance and disappearance of the text content in consecutive frames, the display time range of the subtitles can be determined. For example, when subtitle text is first detected in a certain frame image, the timestamp of this frame image is recorded as the start time of the subtitle. When the subtitle text disappears in subsequent frame images, the timestamp of this frame image is recorded as the end time of the subtitle. Finally, the extracted subtitle text and its corresponding time range can be sorted into the required format, such as output as subtitles = [{'text': '你到底想怎么样?','start': 16.0, 'end': 18.5},...], thus completing the subtitle extraction task and obtaining subtitle information.

[0033] Character recognition in video data can be performed for example by using a multi-task cascaded convolutional neural network (MTCNN) model to detect faces in each frame of the video data, extracting key facial feature points and the location information of face bounding boxes. The extracted information can then be input into an ArcFace model for feature extraction and similarity calculation. Faces with high similarity (e.g., similarity greater than a similarity threshold) are grouped into the same face cluster, forming an initial set of face clusters [face_cluster_1, face_cluster_2, ...]. When subtitle information is detected (e.g., “Li Lei: Stop!”) and a face from a particular face cluster (e.g., face_cluster_1) is prominent, that face cluster can be associated with the character name (“Li Lei”) in the subtitle. In this way, the correspondence between face clusters and character names can be improved, enabling accurate identification and labeling of characters appearing in the video data, thereby obtaining the character information of the video data.

[0034] Scene detection in video data can be performed by analyzing the visual, audio, and content similarities between consecutive shots to identify transition points in the story's location, time, or plot unit, thus segmenting the video data by shot. Keyframes can then be extracted from each shot, selecting the most representative frames to reflect the main content. A ResNet residual network model can then be used to extract features from these keyframes, obtaining feature vectors for each shot. Next, the feature vectors of consecutive shots can be clustered or segmented. Similar shot feature vectors can be grouped together; alternatively, a distance-based sliding window segmentation method can be used to calculate the distance between feature vectors of consecutive shots. When the distance exceeds a set threshold, a significant scene change can be considered. Especially when the feature vectors of consecutive shots show significant drift, such as a complete change in background from indoors to outdoors, this point can be marked as a scene transition point, and its timestamp recorded. Finally, the timestamps of all scene transition points can be compiled into a list, for example, scene_timestamps = [0.0, 350.5, 780.2, ...], thus completing the scene detection task for the video data and obtaining a list of scene transition timestamps.

[0035] S102: Based on the video structure information, divide the target video into at least two content levels.

[0036] Computing devices can divide a target video into at least two content levels based on video structure information. Each content level can include multiple video segments. For example, consider dividing the target video into a minimum content level and a maximum content level. The minimum content level is the shot level, composed of multiple shot segments, each corresponding to a video segment between adjacent shot transition points. The maximum content level is the scene level, composed of multiple scenes, each formed by aggregating consecutive shot segments, where scene transition points are determined by a list of scene transition timestamps. This division is based on the video's narrative structure, ensuring that each content level includes multiple video segments, thereby supporting subsequent multi-level summarization generation.

[0037] S103: Starting from the lowest content level, generate summary text for each content level level step by step.

[0038] The computing device generates summary text for each content level, starting from the lowest content level. The summary text for higher content levels is generated by fusing the summary text of the lower content levels below it.

[0039] For example, a computing device can divide a target video (taking one episode of video as an example) into multiple shot segments based on a list of timestamps of shot transitions, and generate a summary text for each shot segment based on the subtitle information and character information in each shot segment.

[0040] For example, taking a single scene clip as an example, a computing device can integrate the subtitle information, character information, and visual descriptions from that clip, and then pass this information as input to a pre-trained summarization generation model. Upon receiving the input, the summarization generation model can generate a concise and descriptive summary text based on this information, summarizing the core content of the scene clip.

[0041] For example, for the first shot (e.g., 0.0s-15.2s), the data input to the summary generation model can include character information (e.g., "Han Meimei"), caption information (empty in this example), and visual descriptions (e.g., "Close-up shot, character Han Meimei running in the rain, expression sad"). The summary generation model can then output descriptive summary text based on this information, such as "Han Meimei running sadly in the rain." In this way, accurate and expressive summary text can be generated for each shot, helping users quickly grasp the key plot points and emotional atmosphere of each shot in the video.

[0042] The computing device can aggregate multiple shot clips to obtain multiple scenes based on the time point of each scene transition in the timestamp list of scene switching. Based on the aggregated shot clips under each scene, the summary text corresponding to each shot clip, and the subtitle information, it can generate scene summary text corresponding to each scene.

[0043] For example, taking a specific scene as an example, the computing device can determine the summary text of the shot clips within that scene, as well as the subtitle information within the scene, and pass this content as input to the summary generation model. Based on these inputs, the summary generation model can generate a more detailed and comprehensive scene summary text to accurately reflect the core events, character interactions, and emotional atmosphere of the scene.

[0044] For example, for the first scenario (e.g., 0.0s - 350.5s), the data input to the summary generation model could include the following: Summary texts corresponding to multiple shot clips (such as "Han Meimei runs sadly in the rain", "Li Lei drives up to Han Meimei and gets out of the car", "Li Lei and Han Meimei have a heated argument in the rain") and subtitles within the scenes.

[0045] The summary generation model can output detailed scene summary text based on this information, for example, as follows: Scene 1: A rainy night on the street. After breaking up with Li Lei over something, Han Meimei runs away in the rain. Li Lei catches up and tries to stop her, but the two have a heated argument, and their relationship deteriorates further. This summary not only covers the main events and character interactions in the scene, but also reflects the emotional atmosphere of the scene, providing users with a comprehensive understanding of the entire scene.

[0046] S104: Input the highest content level summary text into the pre-trained summary generation model to generate the plot summary text of the target video.

[0047] The computing device can input the highest content-level summary text into a pre-trained summary generation model to generate the plot summary text of the target video. In other words, the summary text of each scene is input into the pre-trained summary generation model to generate the plot summary text of the target video.

[0048] For example, if the target video is a specific episode of a target TV series, which may include multiple episodes, then the summary text of each scene corresponding to that episode can be input into a pre-trained summary generation model to generate a plot summary text for that episode, thereby highlighting the main plot twists and character development of that episode.

[0049] For example, the scene summary text for this episode could be: "Scene 1: Rainy night on the street. Han Meimei breaks up with Li Lei over something and runs away in the rain. Li Lei catches up and tries to stop her, but the two have a heated argument, and their relationship deteriorates further.", "Scene 2: Li Lei returns to the company and feels immense pressure facing a mountain of work. His colleague Xiao Wang comes to comfort him, but Li Lei is distracted.", and "Scene 3: Han Meimei cries alone at home, recalling her memories with Li Lei, filled with contradictions and entanglements." The generated summary text for this episode would be: "This episode tells the story of the emotional crisis between Han Meimei and Li Lei. On a rainy night, Han Meimei breaks up with Li Lei over something and runs away. Li Lei catches up and tries to stop her, but the two have an argument, and their relationship deteriorates. Li Lei returns to the company, distracted by work pressure, and his colleague Xiao Wang's comfort cannot lift him out of his emotional slump. Meanwhile, Han Meimei recalls her past with Li Lei at home, feeling conflicted and entangled. This episode showcases the breakdown of their relationship and their respective struggles in life." In some possible implementations, after generating summary text for each episode of the TV series, embodiments of this application may also analyze the entire series to obtain a plot summary text for the series.

[0050] For example, a computing device can take as input the summary text of each episode of the TV series (from episode_01_summary to episode_40_summary). These summary texts already cover the main plot twists, character development, and key events of each episode. By integrating these summary texts and utilizing a summary generation model (such as a large language model), a complete summary text of the entire series can be generated. This overview highlights the core storyline of the entire series, the evolution of the main character relationships, and the development of key plot points, helping viewers quickly grasp the overall picture of the series. For example, the summary text of the entire series output by the model could be: "This drama is a TV series with the love story of Han Meimei and Li Lei as the main storyline. The series consists of 40 episodes, telling the emotional journey of the two from meeting, falling in love, breaking up, and finally reconciling. The series interweaves multiple subplots such as workplace challenges, family conflicts, and personal growth, showcasing the struggles and choices of modern urbanites in love and career. The first episode opens with an argument on a rainy night, revealing a rift in Han Meimei and Li Lei's relationship. In the subsequent episodes, the two experience many ups and downs in their respective lives. Li Lei faces enormous work pressure at the company, while Han Meimei wavers between family and career. After a series of misunderstandings and conflicts, the two finally reconcile in the fortieth episode, bringing this love story to a happy ending." In this embodiment, the summary generation model acts as a central processing unit, used to fuse and understand heterogeneous data such as visual content, character identities, and text captions. The summary generation model is not only a text generator but also a reasoning engine for narrative logic. Through input instructions, it is guided to intelligently organize fragmented information such as who, when, where, what did, and what said into a storyline with causal relationships and emotional nuances, thereby solving the semantic gap problem and achieving a qualitative leap from data to narrative.

[0051] S105. Input the plot summary text into the summary generation model. Based on the target instruction to generate character relationship information, the summary generation model outputs the character relationship information between characters in the target video.

[0052] The computing device can input plot summary text into the summary generation model, and based on the target instruction to generate character relationship information, the summary generation model outputs the character relationship information between characters in the target video. The character relationship information includes at least the type of relationship between the characters and the temporal change information of the relationship type as the plot progresses.

[0053] For example, a computing device can input a target instruction (here called the first target instruction) and the summary text of multiple target videos (taking any episode of a target TV series as an example) into a summary generation model. The first target instruction instructs the summary generation model to analyze and construct a relationship graph between target characters (e.g., "Li Lei," "Han Meimei," and "Lin Tao") based on the summary text of the multiple videos. The multiple target videos, for example, can be parts of a target TV series, or the graph output by the summary generation model can be presented in JSON format, clearly showing the types of relationships between characters and their evolution.

[0054] Specifically, the plot summary text and the first target instruction are input into the summary generation model; wherein, the first target instruction is used to instruct the summary generation model to perform the following operations: Identify the target characters in the plot summary text, determine the interaction events between different target characters, and determine the episode time point, relationship status, and relationship type corresponding to each interaction event. Based on the interaction events and the corresponding episode time points, organize the relationship status and relationship type according to the plot progression order to generate temporal change information of the relationship status.

[0055] Based on the target roles and the time-series change sequence, a relationship graph is constructed, wherein the edges of the relationship graph are used to represent the relationship information between the roles. The relationship information includes the relationship type between the target roles and the time-series change information. Finally, the relationship graph is output by the summary generation model.

[0056] In other words, the summary generation model first processes the plot summary text to identify the target characters (i.e., the main characters involved in the plot) and determine the interaction events (such as dialogue, conflict, or cooperation) between these target characters. For each interaction event, the summary generation model further extracts its corresponding episode time point (i.e., the timestamp of the event in the video), relationship state (referring to the specific manifestation under a particular relationship type, such as "passionate love," "cold war," and "breakup" in the "love" relationship type), and relationship type (such as abstract relationship categories like family, friendship, and love). Subsequently, based on all interaction events and their episode time points, the summary generation model organizes the relationship states and relationship types according to the chronological order of the plot, generating temporal change information of the relationship states (i.e., describing the process sequence of the relationship state as the plot evolves, such as the dynamic change from "acquaintance" to "friend" and then to "enemy"). Next, the summary generation model constructs a relationship graph based on the identified target roles and the aforementioned temporal change information. The edges of the relationship graph are used to represent the relationship information between roles, including relationship type and temporal change information. Finally, the summary generation model outputs the relationship graph.

[0057] Furthermore, the "nodes" section of the relationship graph lists each character and their description. For example, "Li Lei" is the CEO of XX Company, impulsive but gradually matures; "Han Meimei" is Li Lei's ex-girlfriend, later becoming a business partner; and "Lin Tao" is Li Lei's business rival and love rival. The "edges" section details the relationships between characters and their evolution. For instance, the relationship between "Li Lei" and "Han Meimei" develops from lovers to adversaries, eventually reconciling, with key events including their argument in the rainy night of episode one, their business competition in episode fifteen, and their eventual reconciliation in episode forty. The relationship between "Li Lei" and "Lin Tao" is that of business rivals and love rivals, with key events including losing a bid in episode five and vying for Han Meimei in episode twenty. This relationship graph directly outputs the core information that users care about most—"who is who, and what happened between them"—not only helping viewers quickly understand the complex relationships between characters but also showcasing the dynamic changes of these relationships within the plot. It provides viewers with a clear perspective on the interactions and emotional entanglements between characters, enhancing the viewing experience.

[0058] In some possible implementations, the computing device can associate temporal change information in the relationship graph with one or more corresponding key event evidences, where the key event evidence includes event descriptions from the plot summary text that reflect changes in the relationship state.

[0059] The output relationship graph can be presented in the form of an interactive webpage or application.

[0060] In response to a trigger operation on an edge in the relationship graph interface, the system jumps to and plays the target video segment corresponding to the timestamp of the key event evidence associated with the edge.

[0061] For example, a computing device can associate the temporal change information of relationship states representing the dynamic evolution of relationships in a relationship graph with one or more key event evidences from the plot summary text. Key event evidence refers to specific event descriptions that directly cause or reflect changes in the relationship state between target characters. The resulting relationship graph can be presented as an interactive interface (e.g., a webpage or application interface). When a user performs a triggering operation (such as clicking) on ​​an edge connecting two target character nodes on this interface, the computing device can respond to this operation and automatically jump to the corresponding segment in the target video and play it based on the video timestamp recorded in the key event evidence associated with that edge.

[0062] For example, when a user clicks on the relationship between "Li Lei" and "Han Meimei" in the relationship graph, the page automatically jumps to the scene at 15 minutes and 20 seconds into episode 1 of the video, which is the segment where the two argue in the rainy night; clicking on the relationship between "Li Lei" and "Lin Tao" will jump to the scene at 23 minutes and 10 seconds into episode 5, which is the segment where Li Lei loses the bid. This direct jump function not only allows users to quickly view the original scenes of key events, but also helps them to more intuitively understand the complex relationships and emotional entanglements between the characters. In this way, users can easily delve into specific plot details from the macro perspective of the relationship graph, thereby obtaining a richer and more immersive viewing experience.

[0063] In some possible implementations, embodiments of this application may also use target instructions and the summary text of multiple target videos (taking any episode of a target TV series as an example) as input to the summary generation model. The multiple target videos may be, for example, parts of an episode or each episode of a target TV series. Using the target instructions (which can be called second target instructions), the summary generation model is guided to extract relevant information about the target character from the summary text of the multiple videos and generate a detailed character biography.

[0064] The computing device can input the generated plot summary text (i.e., the complete narrative description obtained after multi-level summarization of the target video content) and a second target instruction into the summary generation model. This second target instruction guides the summary generation model to perform the following operations: First, identify the target characters mentioned in the plot summary text (i.e., the main characters involved in the plot); second, determine the static attribute information of these target characters in the plot (including at least one of identity attributes such as family background and social relationships, and occupational attributes such as job position); next, analyze the behavioral pattern characteristics (referring to the character's recurring behaviors in specific situations) and personality traits (such as courage, suspicion, etc.) exhibited by the target characters during the plot's progression, and identify at least one key event associated with the target character (i.e., events that reflect the target character's personal development trajectory or changes in relationships with other characters), organizing these key events according to the chronological order of the plot's development to form a key event sequence. Finally, based on the static attribute information, behavioral pattern characteristics, personality traits, and key event sequence, a comprehensive character portrait of the target character (a comprehensive description covering the character's basic attributes, behavioral characteristics, personality evolution, and key experiences) is generated, and the summary generation model outputs one or more character portraits of the target characters.

[0065] For example, the target instruction could be: "Based on the summary text of the following 40 episodes, extract all relevant information about the character 'Li Lei' and generate a character biography." The abstract generation model output could be: "Li Lei: The male protagonist of the series, CEO of XX Company. Initially impulsive and domineering, he gradually matures through his emotional entanglement with Han Meimei, learning understanding and responsibility. Key events include: chasing after Han Meimei in the rain, turning the tide during a company crisis, and finally apologizing to Han Meimei, etc..." The character profile generated by the abstract generation model not only covers Li Lei's basic identity information (such as CEO of XX Company), but also delves into his personality traits (impulsiveness and domineering nature in the early stages) and his growth trajectory in the plot (gradually learning to understand and take responsibility amidst his emotional entanglement with Han Meimei). Furthermore, the abstract generation model lists Li Lei's key events, such as chasing after Han Meimei in the rain, turning the tide during the company crisis, and ultimately apologizing to Han Meimei, thus presenting the audience with a three-dimensional, rich, and profound character. The effect of this character profile generation is to provide the audience with a comprehensive and in-depth understanding of the character. By detailing the character's basic identity information, personality traits, growth trajectory, and key events, the audience can quickly and profoundly grasp the character's core qualities and important experiences in the story. This three-dimensional, rich, and profound character image not only enhances the audience's resonance and emotional investment in the character, but also improves their overall understanding and appreciation of the plot, making the character no longer a simple symbol, but a flesh-and-blood person with complex emotions and a growth process.

[0066] In some possible implementations, embodiments of this application may also incorporate audio analysis technology to identify the emotional information of background music (such as excitement, sadness, and suspense), and input this emotional information as an important dimension into the summary generation model. When generating summary texts for clips, scenes, episodes, and the entire series, the summary generation model can incorporate this emotional information, ensuring that the generated descriptions not only cover the plot content but also reflect the emotional atmosphere conveyed by the music, thus making the summaries more impactful and immersive. For example, in a key plot point of a suspense drama, if the background music is tense string music, the summary generation model might emphasize the tension and uncertainty of the scene in the generated summary text; while in a romantic scene, if the background music is gentle piano music, the summary text would highlight warmth and romance. This descriptive approach, combined with audio sentiment analysis, can more comprehensively convey multi-dimensional information about the plot, enhancing the audience's understanding and emotional resonance with the story.

[0067] In some possible implementations, the summary generation model can be trained in the following ways: The summary generation model can use plot summary text including character relationship annotations as training samples. It calculates the difference between the character relationship information generated by the summary generation model and the character relationship annotations in the training samples, and iterates based on the difference values. The character relationship annotations at least include the relationship type between characters and the temporal changes in the relationship type as the plot progresses.

[0068] For example, training samples can consist of plot summary text and corresponding character relationship annotations. The plot summary text refers to a complete narrative description of the target video content generated through the aforementioned multi-level summarization process. Character relationship annotations are manually or semi-automatically labeled data, including at least the relationship type between characters (such as abstract relationship categories like kinship, friendship, and love) and information on the temporal changes in relationship types as the plot progresses (i.e., a sequence describing the evolution of relationship states under a specific relationship type, such as a dynamic sequence from acquaintance to friendship to enmity). During training, these labeled samples can be used to guide the model in learning to accurately extract character relationship information from the plot summary text.

[0069] The internal structure of the summary generation model (hereinafter referred to as the model) can be implemented based on mature large language model (LLM) architectures, such as Transformer base models like BERT (Bidirectional Encoder Representations from Transformers). These models can capture long-distance dependencies in text through self-attention mechanisms, making them suitable for handling narrative content. During training, the model receives plot summary text as input and outputs predicted character relationship information. By comparing the model output with the character relationship annotations in the training samples, the difference value is calculated (e.g., using the cross-entropy loss function or a custom sequence loss function to measure the error in relationship classification and temporal order), and the model parameters are iteratively updated based on the difference value using the backpropagation algorithm until the model converges.

[0070] During training, the guidance can be designed as follows: Based on the given plot summary text, identify the types of relationships between target characters and organize the sequence of changes in these relationships according to the chronological order of the plot. In actual model application, the guidance appears as objective instructions, such as the second objective instruction: Identify the target characters in the plot summary text, extract their relationship information, and generate a chronological sequence of changes. This guidance ensures that the model output is consistent with the annotation format, guaranteeing training effectiveness.

[0071] This application processes target videos to obtain their structural information (such as shots, scene boundaries, subtitles, and character identifiers), establishing a unified multimodal data foundation for subsequent analysis and ensuring the comprehensiveness of information sources. Subsequently, based on this video structural information, the target video is divided into at least two content levels (shots, scenes, etc.) (each level includes multiple video segments), and hierarchical summary text generation is performed from the lowest content level (shots) to higher content levels (scenes, episodes, etc.). In this process, each higher-level summary text is generated by integrating the semantics of its subordinate lower-level summary texts. This fundamentally achieves deep semantic integration of multimodal information such as visual content, character dynamics, and dialogue text, effectively overcoming the problem of information silos. Then, the highest-level content summary text (i.e., the plot summary text), containing complete narrative logic, is input into a pre-trained summary generation model, transforming the structured information after hierarchical fusion into a coherent and concise plot summary text, completing the qualitative change from fragmented facts to a holistic narrative. Finally, this plot summary text is input into the same summary generation model again, and the target instruction to generate character relationship information is used. Based on a deep semantic understanding of the overall plot, the model can infer and output character relationship information that includes not only the type of relationship between characters, but more importantly, the temporal change information of the relationship type as the plot progresses. This achieves accurate construction of dynamic character relationships and improves the user experience.

[0072] To implement the above embodiments, this disclosure also proposes a device for constructing character relationships in videos.

[0073] Figure 2 This is a schematic diagram of a video character relationship construction device provided in an embodiment of this disclosure. The device can be implemented by software and / or hardware, and is generally integrated into a computing device. Figure 2 As shown, the device includes: Data processing unit 210 is used to process the target video to obtain the video structure information of the target video; The segmentation unit 220 is used to divide the target video into at least two content levels according to the video structure information, wherein each content level includes multiple video segments; The first generation unit 230 is used to generate summary text for each content level step by step, starting from the lowest content level, wherein the summary text of a higher content level is generated by fusing the summary text of the lower content levels below it. The second generation unit 240 is used to input the highest content level summary text into a pre-trained summary generation model to generate the plot summary text of the target video; The output unit 250 is used to input the plot summary text into the summary generation model, and output the character relationship information between the characters in the target video based on the target instruction to generate character relationship information. The character relationship information includes at least the relationship type between the characters and the temporal change information of the relationship type as the plot progresses.

[0074] Optional, output unit 250, specifically used for: The plot summary text and the first target instruction are input into the summary generation model; wherein, the first target instruction is used to instruct the summary generation model to perform the following operations: Identify the target character in the plot summary text; Identify the interaction events between different target characters, and determine the corresponding episode time point, relationship status, and relationship type for each interaction event; Based on the interactive events and the corresponding episode time points, the relationship states and relationship types are organized according to the plot progression order to generate temporal change information of the relationship states; Based on the target roles and the temporal change sequence, a relationship graph is constructed, wherein the edges of the relationship graph are used to represent the relationship information between the roles, and the relationship information includes the relationship type between the target roles and the temporal change information; The relationship graph is output from the summary generation model.

[0075] Optionally, the device further includes: The association unit is used to associate the temporal change information in the relationship graph with one or more corresponding key event evidences, wherein the key event evidences include event descriptions from the plot summary text that reflect the change in relationship status.

[0076] Optionally, the device further includes: A presentation unit is used to present the relationship graph; In response to a trigger operation on an edge in the relationship graph interface, the system jumps to and plays the target video segment corresponding to the key event evidence associated with the edge, based on the timestamp.

[0077] Optionally, the device further includes: The input unit is used to input the plot summary text and the second target instruction into the summary generation model, wherein the first target instruction is used to instruct the summary generation model to perform the following operations: Identify the target character in the plot summary text; Determine the static attribute information of the target character in the storyline, including at least one of identity attributes and occupation attributes; Determine the behavioral patterns and personality traits exhibited by the target character during the plot progression; Identify at least one key event associated with the target character, and organize the at least one key event in the order of plot development to form a key event sequence. The key event is an event description that reflects the character development or relationship status change of the target character. Based on static attribute information, behavioral pattern characteristics, personality traits, and key event sequences, a character profile of the target character is generated. The summary generation model outputs character profiles for one or more target characters.

[0078] Optionally, the summary generation model uses plot summary text including character relationship annotations as training samples, calculates the difference value by comparing the character relationship information generated by the summary generation model with the character relationship annotations in the training samples, and iteratively trains based on the difference value; wherein, the character relationship annotation includes at least the relationship type between characters and the temporal change information of the relationship type as the plot progresses.

[0079] Optionally, the video structure information includes a list of timestamps for shot transitions, subtitle information, character information, and a list of timestamps for scene transitions in the target video; The first generation unit is specifically used for: Based on the timestamp list of the shot transitions, multiple shot segments are determined from the target video, and a summary text corresponding to each shot segment is generated based on the subtitle information and character information corresponding to each shot segment. Based on the timestamp list of scene switching, the multiple shot clips are aggregated to obtain multiple scenes. Based on the aggregated shot clips under each scene, the summary text corresponding to each shot clip, and the subtitle information, a scene summary text corresponding to each scene is generated. Each scene summary text is input into a pre-trained summary generation model to generate a plot summary text for the target video.

[0080] Optionally, the first generating unit is specifically used for: Based on the subtitle information, character information, and emotional information of the background music in each shot, a summary text for each shot is generated.

[0081] The video character relationship construction apparatus provided in this disclosure can execute the video character relationship construction method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0082] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the video character relationship construction method in the above embodiments.

[0083] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of the present disclosure.

[0084] The following is a detailed reference. Figure 3 The diagram illustrates a structural schematic suitable for implementing the computing device 300 in the embodiments of this disclosure. The computing device 300 in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The computing device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0085] like Figure 3 As shown, the computing device 300 may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the computing device 300. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0086] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows computing device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 A computing device 300 with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0087] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a memory 308, or installed from a ROM 302. When the computer program is executed by the processor 301, it performs the functions defined in the video character relationship construction method of embodiments of this disclosure.

[0088] It should be noted that the computer-readable medium described above in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium that includes or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code included on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0089] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0090] The aforementioned computer-readable medium may be included in the aforementioned computing device; or it may exist independently and not assembled into the computing device.

[0091] The aforementioned computer-readable medium carries one or more programs, which, when executed by the computing device, cause the computing device to perform the method for constructing character relationships in the aforementioned video.

[0092] The computing device can be programmed with computer program code in one or more programming languages ​​or a combination thereof to perform the operations of this disclosure. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0093] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which includes one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0094] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0095] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0096] In the context of this disclosure, a machine-readable medium can be a tangible medium that may include or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0097] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0098] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0099] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for constructing character relationships in a video, characterized in that, include: Data processing is performed on the target video to obtain the video structure information of the target video; Based on the video structure information, the target video is divided into at least two content levels, wherein each content level includes multiple video segments; Starting from the lowest content level, summary text for each content level is generated progressively. The summary text for higher content levels is generated by fusing the summary texts of the lower content levels below. The highest content-level summary text is input into a pre-trained summary generation model to generate a plot summary text for the target video; The plot summary text is input into the summary generation model. Based on the target instruction to generate character relationship information, the summary generation model outputs the character relationship information between the characters in the target video. The character relationship information includes at least the relationship type between the characters and the temporal change information of the relationship type as the plot progresses.

2. The method according to claim 1, characterized in that, The step of inputting the plot summary text into the summary generation model, and based on the target instruction to generate character relationship information, having the summary generation model output the character relationship information between characters in the target video, includes: The plot summary text and the first target instruction are input into the summary generation model; wherein, the first target instruction is used to instruct the summary generation model to perform the following operations: Identify the target character in the plot summary text; Identify the interaction events between different target characters, and determine the corresponding episode time point, relationship status, and relationship type for each interaction event; Based on the interactive events and the corresponding episode time points, the relationship states and relationship types are organized according to the plot progression order to generate temporal change information of the relationship states; Based on the target roles and the temporal change sequence, a relationship graph is constructed, wherein the edges of the relationship graph are used to represent the relationship information between the roles, and the relationship information includes the relationship type between the target roles and the temporal change information; The relationship graph is output from the summary generation model.

3. The method according to claim 2, characterized in that, The method further includes: The temporal change information in the relationship graph is associated with one or more corresponding key event evidences, including event descriptions from the plot summary text that reflect changes in the relationship status.

4. The method according to claim 3, characterized in that, The method further includes: Present the aforementioned relationship graph; In response to a trigger operation on an edge in the relationship graph interface, the system jumps to and plays the target video segment corresponding to the key event evidence associated with the edge, based on the timestamp.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: The plot summary text and the second target instruction are input into the summary generation model, wherein the first target instruction is used to instruct the summary generation model to perform the following operations: Identify the target character in the plot summary text; Determine the static attribute information of the target character in the storyline, including at least one of identity attributes and occupation attributes; Determine the behavioral patterns and personality traits exhibited by the target character during the plot progression; Identify at least one key event associated with the target character, and organize the at least one key event in the order of plot development to form a key event sequence. The key event is an event description that reflects the character development or relationship status change of the target character. Based on static attribute information, behavioral pattern characteristics, personality traits, and key event sequences, a character profile of the target character is generated. The summary generation model outputs character profiles for one or more target characters.

6. The method according to claim 1, characterized in that, The summary generation model uses plot summary text including character relationship annotations as training samples. It calculates the difference value by comparing the character relationship information generated by the summary generation model with the character relationship annotations in the training samples, and iterates and trains based on the difference value. The character relationship annotations include at least the relationship type between characters and the temporal change information of the relationship type as the plot progresses.

7. The method according to claim 1, characterized in that, The video structure information includes a list of timestamps for shot transitions, subtitle information, character information, and a list of timestamps for scene transitions in the target video. Based on the timestamp list of the shot transitions, multiple shot segments are determined from the target video, and a summary text corresponding to each shot segment is generated based on the subtitle information and character information corresponding to each shot segment. Based on the timestamp list of scene switching, the multiple shot clips are aggregated to obtain multiple scenes. Based on the aggregated shot clips under each scene, the summary text corresponding to each shot clip, and the subtitle information, a scene summary text corresponding to each scene is generated. Each scene summary text is input into a pre-trained summary generation model to generate a plot summary text for the target video.

8. The method according to claim 7, characterized in that, The step of generating a summary text for each shot segment based on the subtitle and character information in each shot segment includes: Based on the subtitle information, character information, and emotional information of the background music in each shot, a summary text for each shot is generated.

9. A computing device, characterized in that, The computing device includes: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the method of any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing the method described in any one of claims 1-8.

Images