An artificial intelligence-based script outline creation system

Abstract

The present application provides a kind of script outline creation system based on artificial intelligence, it includes: information identification module is through the preliminary analysis of the identification of the attribute of the information of the person, object and background in picture in convolutional neural network;Comprehensive analysis module is the in-depth analysis of the story type, main character and supporting actor and their interests of script outline;Result output module is through the word coding of the relevant information of picture analysis by recurrent neural network, then the relevant words are decoded and output script outline.The above-mentioned method applies artificial intelligence technology to the field of script outline creation, so as to assist the creator to create or provide them with inspiration.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of script outline creation, and specifically relates to a script outline creation system based on artificial intelligence. BACKGROUND

[0002] A script outline is a brief outline at the early stage of script creation, and is a brief summary of the plot. It has a guiding role for later script creation. The creator will further refine the script outline to perfect the story framework into a script containing rich details such as scenes, actions, and dialogues. The writing of a script outline mainly has two forms: artificial and semi-artificial. The traditional artificial method is usually the idea generated by the creator being touched by his own experience or what he has heard. This process relies on the inspiration and idea of the creator. The semi-artificial method is to use a certain algorithm to combine and compose the story by pre-collected stories, focusing on the language organization at the text level.

[0003] Patent: CN112464060A discloses a method and system for assisting script writing based on Wiener's law. Based on the operation principle of the cyber chain in Wiener's law, combined with the classification and connection method of elements such as plot, scene, character, logic, and target in script creation work, relevant knowledge and tools in the computer field are used to effectively process various information in script production work to obtain the corresponding script prototype. The auxiliary method of this patent still focuses on the form of text, and still needs the participation of the creator.

[0004] The disadvantages of the prior art are that the artificial or semi-artificial script outline writing method has low efficiency and potential infringement risk. The key to artificial script outline writing is the writing state of the creator, which is difficult to guarantee the stability of the creation. The existing semi-artificial method needs to store a large amount of story materials and create on the basis of others' stories, which may have potential risk of infringement.

[0005] The technical problem to be solved by the present application is to provide a script outline creation system based on artificial intelligence, which generates a corresponding script outline by extracting and analyzing a series of input pictures, thereby assisting the creator in creation or providing them with creative inspiration. SUMMARY

[0006] The present application preliminarily analyzes and deeply analyzes the relevant information in the pictures through a convolutional neural network, then encodes the relevant information of the pictures through a recurrent neural network, and finally decodes and outputs the story type, story abstract, character biography, and story introduction of the script outline based on the above work. It is specifically divided into the following modules:

[0007] The information recognition module is a preliminary analysis of the information of the characters, objects and backgrounds in the picture and their attributes through a convolutional neural network.

[0008] The comprehensive analysis module is an in-depth analysis of the story type, main characters and supporting characters and their relationships in the script outline.

[0009] The result output module is a script outline output through a recurrent neural network that first encodes the relevant information of the picture analysis into words and then decodes the relevant words.

[0010] Further, the specific process of the picture information preliminary analysis method is as follows:

[0011] First, the user input picture is preprocessed, and then the information of the characters, objects and backgrounds in the picture and their attributes are recognized through a convolutional neural network.

[0012] Specifically, the input picture preprocessing method is as follows:

[0013] The input picture is preprocessed by light compensation, gray scale transformation, histogram equalization, normalization, geometric correction, filtering and sharpening.

[0014] Specifically, the method for recognizing the information of the characters in the picture is as follows:

[0015] The characters and their related information in the picture are recognized. First, the number of characters in the picture is determined by face detection, and then the information of each character and their related information in the picture is analyzed, including the gender, age, face shape, eye shape, nose shape, mouth shape, expression and action of the character, and each character and their related information is recorded in the relevant information database of the corresponding picture.

[0016] Specifically, the method for recognizing the information of the objects in the picture is as follows:

[0017] The objects and their related information in the picture are recognized. First, the objects in the picture are recognized, and then the information of each object and their related information in the picture is analyzed, including the shape, color, texture and position relationship of the object, and each object and their related information is recorded in the relevant information database of the corresponding picture.

[0018] Specifically, the method for recognizing the information of the backgrounds in the picture is as follows:

[0019] The background and its related information in the picture are recognized. First, the background in the picture is recognized, and then the related information of the background in the picture is analyzed, including the scene, time and time, and each object and their related information is recorded in the relevant information database of the corresponding picture.

[0020] Further, the specific process of the picture information deep analysis is as follows:

[0021] First, the role allocation of the script outline is analyzed, then the type of the script outline is analyzed, and then the main character and the supporting character are determined according to the role allocation and the type of the script outline, and finally the relationship between the roles in the script outline is analyzed.

[0022] Specifically, the method for analyzing the role allocation of the script outline is:

[0023] First, the information library of the same character is merged according to the gender, age, eye type, nose type, mouth type and face type of the character in each picture, and the number of characters appearing in the story is determined. Then, the characters are sorted from high to low according to the number of appearances, and the top two male characters and the top two female characters are selected as the candidate characters of the main character and the main supporting character (if the number of appearances of the candidate characters is consistent, further filtering their appearance pictures is performed, and the character in the center of the picture is preferred), and the characters whose appearance frequency is less than 30% of the total number of user input pictures are ignored, and the remaining characters are defined as secondary supporting characters.

[0024] Specifically, the method for analyzing the type of the script outline is:

[0025] According to the preliminary analysis results of the picture information, the characters, object information and background information of all pictures that meet the conditions of a certain story type and have an appearance frequency greater than 50% are determined as the story type.

[0026] Specifically, the method for determining the main character and the supporting character of the script outline is:

[0027] According to the analysis results of the type of the script outline, further analysis is performed from the perspective of different story types, and the main character and the supporting character of the script outline are determined according to the characteristics of the type.

[0028] Specifically, the method for analyzing the relationship between the roles in the script outline is:

[0029] The picture information library of the main character and other characters appearing together is analyzed to determine the relationship between the two characters. If the expression and action information between the two characters has negative emotions such as anger and disgust, or contentious actions such as arguing and fighting, and the appearance frequency is greater than 50%, it is determined that the relationship between the two characters is hostile; otherwise, it is determined that the relationship between the two characters is friendly.

[0030] Further, the specific process of outputting the script outline result is:

[0031] First, the related information of the picture analysis is encoded by a recurrent neural network, and then the related words are decoded to output the story type, character biography and story introduction of the script outline, and finally the story summary is generated by the way of extractive text summary.

[0032] Specifically, the method for encoding the relevant information of picture analysis by words is:

[0033] First, all n words are one-hot encoded, that is, represented by a vector Ii(i=1, 2, 3, …, n) of length n, and the vector Ii has only one index bit with a value of 1 and the rest of the bits are 0. Then the one-hot encoding is vectorized by word2vec to obtain the word vector distribution representation Vi(i=1, 2, 3, …, n) of n words.

[0034] Specifically, the method for outputting the script outline story type is:

[0035] According to the analysis result of in-depth analysis of the type to which the script outline belongs, the story type result determined by the system is output.

[0036] Specifically, the method for outputting the script outline character biography is:

[0037] First, according to the analysis result of in-depth analysis of the main characters and main supporting characters and their relationships, a relevant picture information database of each character is established, which includes gender information, age information, and character relationships, as well as action information, scene information, time information, and time information of the character in each picture. Then, the gender information, age information, relationship, action information, and scene information of the character are decoded by word encoding, that is, each word encoding is input into LSTM, and then the LSTM hidden layer content is input into the softmax function through a fully connected network for prediction, outputting the probability of the predicted word. Finally, the word with the maximum probability value is output according to the prediction probability, and the character biography of each character is obtained.

[0038] Specifically, the method for outputting the script outline story introduction is:

[0039] First, a relevant information database of each picture is established, which includes information of each character and their expression information and action information, object information, and background information in each picture. Then, the character information, object information, and scene information of each picture are decoded by word encoding, that is, each word encoding is input into LSTM, and then the LSTM hidden layer content is input into the softmax function through a fully connected network for prediction, outputting the probability of the predicted word. Finally, the word with the maximum probability value is output according to the prediction probability, and the sentence description of the story introduction is obtained.

[0040] Specifically, the method for outputting the script outline story summary is:

[0041] First, the sentences of the story introduction are preprocessed by word segmentation and stop word filtering, then the word frequency statistics of the preprocessed words are obtained to obtain the key words, and the words are valued according to the word frequency statistics. Then, the scores of the key words contained in the sentence description of each picture are calculated by combining the sentence description of each picture, the word frequency statistics and the scores, and the sentences with the top 10 scores are selected as the story abstract. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0043] Figure 1 is a module composition diagram provided by the embodiment of the present application. DETAILED DESCRIPTION

[0044] The technical solutions of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] Embodiment one

[0046] Referring to Figure 1 The artificial intelligence-based script outline creation system of the present embodiment includes the following modules:

[0047] The first module is an information recognition module, which performs preliminary analysis on the information of the characters, objects and background in the picture and their attributes through a convolutional neural network;

[0048] The second module is a comprehensive analysis module, which performs in-depth analysis on the story type, main characters and supporting characters and their interests of the script outline;

[0049] The third module is a result output module, which first encodes the relevant information of the picture analysis through a recurrent neural network, and then decodes the relevant words to output the script outline.

[0050] Embodiment two

[0051] At the specific implementation level, the present embodiment provides a method for preliminary analysis of picture information, which is as follows:

[0052] Step (1): Preprocessing the num pictures inputted by user, mainly including light compensation, gray scale transformation, histogram equalization, normalization, geometric correction, filtering and sharpening.

[0053] Step (2): Recognizing the person and its related information in the picture. First, face detection is performed to determine the number of persons in the picture, and then face recognition is performed to analyze each person and its related information in the picture, wherein the related information includes the gender, age, face shape, eye type, nose type, mouth type, expression and action of the person. The face shape information includes goose egg face, square face, diamond face, triangular face, heart-shaped face, round face, long face, etc.; the eye type information includes apricot eye, lychee eye, Danfeng eye, Rui Feng eye, squint eye, triangular eye, peach blossom eye, willow leaf eye, hanging eye, etc.; the nose type information includes wide nose bridge, hawk nose, upturned nose, tilted nose, garlic nose, collapsed beam nose, etc.; the mouth type information includes thick lips, thin lips, heart-shaped lips, smiling lips, drooping lips, etc.; the expression information includes happy, sad, angry, fearful, surprised, disgusted, neutral; the action information includes standing, sitting, walking, running, jumping, etc. The related information of each person is recorded in the related information database of the first picture, figure1 1 = {gender, age, face, eye, mouth, expression, action}, figure1 2 = {gender, age, face, eye, mouth, expression, action}, figure1 3 = {gender, age, face, eye, mouth, expression, action}, … (wherein subscript 1 represents the first picture)

[0054] Step (3): Recognizing the object and its related information in the picture. First, object recognition (animals, plants, various props, etc.) is performed on the picture, and then the related information of each object in the picture is analyzed, wherein the related information includes the shape, color, texture and location relationship of the object. The shape information includes circle, semicircle, square, triangle, rectangle, diamond, pentagon, hexagon, heptagon, cone, cylinder, trapezoid, parallelogram; the color information includes red, orange, yellow, green, cyan, blue, purple, gray, pink, black, white, brown; the texture information includes various natural textures, artificial textures, mixed textures; the location relationship includes up, down, left, right, front, back. The related information of each object is recorded in the related information database of the first picture, object1 1 = {shape, color, texture, location}, object1 2={shape, color, texture, location}, object1 3 ={shape, color, texture, location}, … (where subscript 1 represents the first picture)

[0055] Step (4): Identify the picture background and its related information. First, identify the background of the picture, and then analyze the related information of the picture background, which includes scene, time and season. The scene information includes school, hospital, library, cinema, etc.; the time information includes morning, afternoon, evening, night, midnight, early morning (through the sky color and time display in the picture); the season information includes spring, summer, autumn and winter. Enter each object and its related information into the related information database of the first picture, background1 1 ={scene, time, season}, background1 2 ={scene, time, season}, background1 3 ={scene, time, season}, … (where subscript 1 represents the first picture)

[0056] Step (5): Identify the character information, object information and background information of other pictures. Repeat steps (2) to (4) to obtain the related information database of other pictures.

[0057] Example Three

[0058] At the specific implementation level, the embodiment provides a method for in-depth analysis of picture information, which is as follows:

[0059] Step (1): Analyze the role assignment of the story outline. First, according to the gender, age, eye type, nose type, mouth type and face type of the characters in each picture, the related information database of the same character is merged to determine the number of characters appearing in the story. Then, according to the statistical results of the number of appearances of each character, from high to low, the male characters ranked in the top two and the female characters ranked in the top two are selected as the candidate characters of the main characters and the main supporting characters (if the number of appearances of the candidate characters is consistent, further filter the pictures they appear in, and prefer the characters who appear more in the center of the picture), and the characters whose appearance frequency and the proportion of the total number of user input pictures are less than 30% are ignored, and the remaining characters are defined as secondary supporting characters.

[0060] Step (2): Analyzing the type of the story outline. According to the analysis results of step (2), step (3) and step (4) of embodiment two, the information of the characters, objects and backgrounds in all the pictures that meet the conditions of a certain story type and have a frequency greater than 50% is determined to be the story type. For example, if the expression information and action information of the main character and the main supporting character in the pictures in which they appear have strong fluctuations of emotions such as happiness and sadness or actions such as holding hands and kissing, Af accumulates one point, and the story is determined to be a love story; if the expression information of the main character and the main supporting character in the pictures in which they appear has fear, Hf accumulates one point, and the story is determined to be a horror story; if the action information of the main character and the main supporting character in the pictures in which they appear has thrilling actions such as fighting and chasing, Am accumulates one point, and the story is determined to be an action story.

[0061] Step (3): Analyzing the main character and the supporting character of the story outline. According to the analysis results of step (2) of embodiment three, the main character and the supporting character are analyzed from the perspective of different story types. For example, for a love story, when the candidate male character and the candidate female character appear in the pictures most frequently in actions such as holding hands, hugging and kissing, they are defined as the male lead and the female lead, and the other two characters are defined as the main male lead and the main female lead; for a horror story, when the candidate male character and the candidate female character appear in the pictures least frequently in fear, they are defined as the male lead and the female lead, and the other two characters are defined as the main male lead and the main female lead; for an action story, when the candidate male character and the candidate female character appear in the pictures most frequently in thrilling actions such as fighting and chasing, they are defined as the male lead and the female lead, and the other two characters are defined as the main male lead and the main female lead.

[0062] Step (4): Analyzing the relationship between the characters of the story outline. The information library related to the pictures in which the main character and other characters appear together is analyzed to determine the relationship between the two characters. If the expression and action information between the two characters has negative emotions such as anger and disgust or contentious actions such as arguing and fighting and the frequency is greater than 50%, it is determined to be an enemy relationship; otherwise, it is determined to be a friend relationship.

[0063] Embodiment four

[0064] At the specific implementation level, the embodiment provides a method for outputting a story outline sentence description, which is as follows:

[0065] Step (1): word coding of the relevant information obtained from the preliminary analysis and in-depth analysis of the above pictures. First, all n words are one-hot coded, that is, represented by a vector Ii (i = 1, 2, 3, …, n) of length n, and the vector Ii has only one index bit with a value of 1, and the rest of the bits are 0. Then the one-hot coding is vectorized using word2vec to obtain the word vector distribution representation Vi (i = 1, 2, 3, …, n) of the n words.

[0066] Step (2): output the story type. According to the analysis results of step (2) of example three, the determined story type result is output.

[0067] Step (3): output the character biography. First, according to the analysis results of steps (3) and (4) of example three, a relevant picture information library of each character is established, which includes not only the gender information, age information and character relationship of the character, but also the action information, scene information, time information and time information of the character in each picture. Then, the gender information, age information, relationship, action information and scene information of the character are decoded according to the time information and time information of each picture in sequence, that is, each word coding is first input into LSTM, then the LSTM hidden layer content is input into the softmax function through a fully connected network for prediction, the probability of the predicted word is output, and finally the word with the maximum probability value is taken as the output according to the prediction probability, obtaining the character biography of each character.

[0068] Step (4): output the story synopsis. First, a relevant information library of each picture is established, which includes the information of each character and its expression and action information, object information and background information in each picture, and then the information of each character, object and scene in each picture is decoded according to the time information and time information of each picture in sequence, that is, each word coding is first input into LSTM, then the LSTM hidden layer content is input into the softmax function through a fully connected network for prediction, the probability of the predicted word is output, and finally the word with the maximum probability value is taken as the output according to the prediction probability, obtaining the sentence description of the story synopsis.

[0069] Step (5): output the story abstract. First, the sentence description of the story synopsis is preprocessed by word segmentation and stop word filtering, and then the word frequency of the preprocessed words is counted to obtain the key words, and the words are valued according to the word frequency statistics. Then, combined with the sentence description and word frequency statistics of each picture and its score, the score of the key words contained in the sentence description of each picture is calculated, and the top 10 sentences with the highest scores are selected as the story abstract.

[0070] Obviously, the above embodiments are merely example for clearly illustrating but not limitation to the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments need not and can not be enumerated. The changes or variations thus induced are still within the protection scope of the present application.

Claims

1.A script outline creation system based on artificial intelligence, characterized by, Comprise: The information recognition module is the preliminary analysis of the information of the characters, objects and background in the picture and their attributes through convolutional neural network; The comprehensive analysis module is the in-depth analysis of the story type, main characters and supporting characters and their interests in the script outline; The result output module is to encode the relevant information of the picture analysis through recurrent neural network, and then decode the relevant words to output the script outline; The script outline result output method comprises the following steps: First, through the recurrent neural network, the relevant information of the picture analysis is encoded, and then the relevant words are decoded to output the story type, character biography and story introduction of the script outline, and finally the story summary is generated by the way of extractive text summary; The method for outputting the story summary of the script outline comprises the following steps: First, the sentence description of the story introduction is preprocessed by word segmentation and stop word filtering, then the keyword is obtained by counting the word frequency of the preprocessed words, and the words are valued according to the word frequency statistics result, then the score of the key words in the sentence description of each picture is calculated by combining the sentence description of each picture, the word frequency statistics result and its score, and the sentences with the top 10 scores are selected as the story summary. 2.The artificial intelligence-based scenario outline authoring system of claim 1, wherein The method for preliminary analysis of picture information comprises: First, the user input picture is preprocessed by light compensation, gray scale transformation, histogram equalization, normalization, geometric correction, filtering and sharpening, and then the information of the characters, objects and background in the picture and their attributes are recognized through convolutional neural network. 3.The artificial intelligence-based scenario outline authoring system of claim 2, wherein, The method for recognizing the information of the picture comprises: Recognize the characters and their related information in the picture, first determine the number of characters by face detection, then analyze each character and its related information in the picture by face recognition, wherein the related information includes the gender, age, face shape, eye shape, nose shape, mouth shape, expression and action of the character, and each character and its related information is recorded in the related information database of the corresponding picture; Recognize the objects and their related information in the picture, first recognize the objects in the picture, then analyze each object and its related information in the picture, wherein the related information includes the shape, color, texture and position relationship of the object, and each object and its related information is recorded in the related information database of the corresponding picture; Recognize the background and its related information of the picture, first recognize the background of the picture, then analyze the related information of the picture background, wherein the related information includes the scene, time and time, and each object and its related information is recorded in the related information database of the corresponding picture. 4.The artificial intelligence-based scenario outline authoring system of claim 1, wherein The method for in-depth analysis of picture information comprises: First, analyze the role allocation of the script outline, then analyze the type of the script outline, then determine the main characters and supporting characters according to the role allocation and script type, and finally analyze the interests between the characters in the script outline. 5.The artificial intelligence-based scenario outline authoring system of claim 4, wherein The method for analyzing the story type, main characters and supporting characters and their interests in the script outline comprises: The analysis of the role assignment of the script outline is as follows: firstly, the information library of the same character is merged according to the gender, age, eye type, nose type, mouth type and face type of the character in each picture, the number of the characters appearing in the story is determined, then the characters are sorted from high to low according to the statistical result of the number of times each character appears, the male character appearing in the top two and the female character appearing in the top two are the candidate characters of the main character and the main supporting actor, if the number of times of the candidate characters is consistent, the pictures in which the candidate characters appear are further screened, the characters appearing in more central positions of the pictures are preferentially selected, the characters appearing less than 30% of the total number of the input pictures are ignored, and the remaining characters are defined as the minor supporting actors; The analysis of the type to which the script outline belongs is as follows: according to the preliminary analysis result of the picture information, the information of the characters, objects and backgrounds in all the pictures is determined as the story type if the information meets the conditions of the story type and the appearance frequency is greater than 50%; The analysis of the main character and the supporting actor of the script outline is as follows: according to the analysis result of the type to which the script outline belongs, the main character and the supporting actor of the script outline are further analyzed from the perspective of different story types, and the main character and the supporting actor are determined according to the characteristics of the story type; The analysis of the relationship between the characters of the script outline is as follows: the picture information library in which the main character and other characters appear together is analyzed to determine the relationship between the two characters, if the expression and action information between the two characters shows negative emotions such as anger and disgust or shows contentious actions such as quarreling and fighting and the appearance frequency is greater than 50%, the relationship between the two characters is determined as an enemy relationship; otherwise, the relationship between the two characters is determined as a friendship. 6.The artificial intelligence-based scenario outline authoring system of claim 1, wherein The method for encoding the picture analysis related information words comprises the following steps: Firstly, all n words are one-hot encoded, that is, a vector Ii with a length of n is used to represent the one-hot encoding, the vector Ii has only one index bit with a value of 1, and the rest of the bits are all 0; then, the one-hot encoding is vectorized by using word2vec to obtain the word vector distribution representation Vi of the n words, wherein i = 1, 2, 3, …, n. 7.The artificial intelligence-based scenario outline authoring system of claim 1, wherein, The method for outputting the character profile of the script outline comprises the following steps: Firstly, the picture information library of each character is established according to the analysis result of the main character and the main supporting actor and the relationship between the characters, wherein the picture information library includes the gender information, age information and relationship between the characters of the character, and also includes the action information, scene information, time information and time point information of the character in each picture; then, the word encoding of the gender information, age information, relationship, action information and scene information of the character is decoded, that is, each word encoding is input into LSTM, then the content of the LSTM hidden layer is input into a softmax function through a full connection network to predict the probability of the output word, finally, the word with the maximum probability value is output as the output according to the prediction probability, and the character profile of each character is obtained. 8.The artificial intelligence-based scenario outline authoring system of claim 1, wherein, The method for outputting the story introduction of the script outline comprises the following steps: First, the relevant information of each picture is established, including the information of each character, expression and action, object and background in each picture, and then the time information and time information of each picture are decoded in sequence, that is, each word coding is input into LSTM, and then the LSTM hidden layer content is input into the softmax function through the full connection network to predict the probability of the predicted word, and finally the word with the maximum probability value is taken as the output according to the predicted probability, and the sentence description of the story introduction is obtained.

Images