TWI687825B - Method and system for mapping from natural language to color combination - Google Patents

Abstract

The present invention provides a method and a system for mapping from natural language to color combination. The method and system include: dividing an inputted description sentence into a plurality of phrases; expanding each of the plurality of phrases into a cognate glossary having a plurality of cognate words by a word embedding module and merging a plurality of cognate glossaries into an overall words expending collection; calculating a word similarity between each cognate word in the overall words expending collection and predefined color-related words and ranking the word similarities with a high and low value order to select at least one color-related word having a higher word similarity; and outputting a color combination which is predefined and corresponded to the selected at least one color-related word.

Description

Method and system corresponding to natural language to color combination

The invention relates to a method and system for obtaining color image vocabulary and color combination from colloquial language correspondence. In detail, it relates to a method and system corresponding to natural language to color combination.

Color enters the human eye through reflection, etc. and connects with the sense. Generally, color has a common and common feeling response to people. In the design field, a color is often used to convey the target image. How to use color matching to communicate the message efficiently Has become a hot and important research topic.

Color is based on personal memory to form the connection between color and image. Although people have experience or cultural differences in color, there are still general and common feelings and reactions. For example, red often reminds people of flames and generates passion Image and green often make people think of nature and produce comfortable images. Each color has different emotional characteristics, called color images.

Regarding color and imagery, there are different interpretation methods in the field of color research, and there are related discussions in existing technologies. The research results are roughly divided into four categories. The first type corresponds to monochromatic color and imagery. For example, US Patent Publication No. US 2006/0248081 A1 "Color selection method and system using semantic color names"; the second type is corresponding to the color or image by the picture, for example, "A method, device and system for matching the color of the probe" of China Invention Patent Publication No. CN 101546343 A, Republic of China Invention Patent Bulletin No. I 534647 "Custom Picture Template System", and Chinese Invention Patent Publication No. CN 107168968 A "Emotion-oriented Image Color Extraction Method and System"; the third category is based on hue Harmonic color matching, for example, "Home decor color matching" of US Patent Publication No. US 2018/0158123 A1; the fourth category is the only database without automation methods, such as "establishment of the Republic of China Invention Patent No. I228669" Method of Image Scale Analysis Table", and "Color Image Prediction System and Method" of the Republic of China Invention Patent Announcement No. I459220.

It can be seen from the above that in the existing technology, there is a lack of an automatic correspondence method corresponding to the color image vocabulary in natural language and further corresponding to the color combination. Therefore, it helps people to quickly find the corresponding color combination from the spoken language. Issues urgently needed to be resolved in this field.

In order to solve the above problems, the present invention provides a method corresponding to natural language to color combination, which includes: a word segmentation module to divide the narrative sentence into multiple words; and embedding each of the multiple words through a word embedding model Expand into similar vocabulary groups with multiple similar vocabularies; aggregate the similar vocabularies of the multiple vocabularies into a total vocabulary expansion set; calculate the vocabulary similarity between each of the similar vocabulary of the total vocabulary expansion set and the preset color image vocabulary ; Sort the vocabulary similarity based on numerical value; select the word At least one color image vocabulary with a higher similarity ranking is output; and at least one selected color combination corresponding to the preset color image vocabulary is output.

As in the aforementioned method, wherein the word segmentation module is a Chinese word segmentation algorithm of Jieba (Jieba).

As in the aforementioned method, the word embedding model is constructed by a shallow, two-layer neural network trained by the word2vec algorithm.

，其中，E(R)為該總詞彙擴展集，r _i 為該多個詞彙，藉由該Word2vec演算法擴展為m組的該相近詞彙組e(r _i )，且共有p個相近詞彙α。 As in the aforementioned method, wherein the total vocabulary expansion set is calculated by the following equation:

, Where E(R) is the total vocabulary expansion set, and r _{i is} the multiple vocabularies, which are expanded into m groups of similar vocabulary groups e ( r _i ) by the Word2vec algorithm, and there are p similar vocabularies α .

，其中，c _k 為該色彩意象詞彙，該總詞彙擴展集中有l個該相近詞彙α _l ，l為從1至p，sim(c _k ,α _l )為c_k與α_l所計算出的詞彙相似度。 The method as described above, wherein the computing of the total extension word set to the close similarity of each vocabulary word with a preset vocabulary of color images, through which the similarity-based computing equation S _k, S _k equation is based on the similarity

, Where c _{k is} the color image vocabulary, the total vocabulary expansion set has l the similar vocabulary α _l , l is from 1 to p , and sim ( c _k ,α _l ) is calculated by c _k and α _l Vocabulary similarity.

As in the aforementioned method, wherein the color combination is an RGB color combination based on hue and tone arrangement combination.

The present invention also provides a system corresponding to natural language to color combination, which includes: a word segmentation module to divide the narrative sentence into multiple words; an expansion module to embed each of the multiple words through word embedding ) The model is expanded into similar vocabulary groups with multiple similar vocabularies, and the similar vocabularies of the multiple vocabulary are combined to form a total vocabulary expansion set; a semantic correspondence module is used to calculate each similar vocabulary and Preset the lexical similarity of the color image vocabulary, and sort the vocabulary similarity based on the numerical value to select at least one of the color image vocabulary with the highest similarity of the vocabulary; and the color construction module to output at least one of the selected The color image vocabulary presets the corresponding color combination.

As in the aforementioned system, wherein the word segmentation module has a Chinese word segmentation algorithm (Jieba).

As in the aforementioned system, wherein the word embedding model of the expansion module is constructed by a shallow, two-layer neural network trained by the word2vec algorithm.

，其中，E(R)為該總詞彙擴展集，r _i 為該多個詞彙，藉由該Word2vec演算法擴展為m組的該相近詞彙組e(r _i )，且共有p個相近詞彙α。 As in the aforementioned system, where the total vocabulary expansion set is calculated by the following equation:

, Where E(R) is the total vocabulary expansion set, and r _{i is} the multiple vocabularies, which are expanded into m groups of similar vocabulary groups e ( r _i ) by the Word2vec algorithm, and there are p similar vocabularies α .

，其中，c _k 為該色彩意象詞彙，該總詞彙擴展集中有l個該相近詞彙α _l ，l為從1至p，sim(c _k ,α _l )為c_k與α_l所計算出的詞彙相似度。 Close to each of the word preceding the system, wherein the system module corresponding to semantic similarity equation S _k by calculating the total words and extended set of the predetermined color image vocabulary word similarity, the similarity equation S _k system

, Where c _{k is} the color image vocabulary, the total vocabulary expansion set has l the similar vocabulary α _l , l is from 1 to p , and sim ( c _k ,α _l ) is calculated by c _k and α _l Vocabulary similarity.

As in the aforementioned system, wherein the color combination of the color construction module is an RGB color combination based on hue and tone arrangement combination written in the Python programming language.

By the method or system of the present invention, the user enters the narrative sentence of the color image vocabulary to be queried, and the color image vocabulary and its color combination of similar meaning can be automatically obtained, and the corresponding RGB values are provided; through the color image vocabulary of the present invention to the color The automatic conversion of the combination can assist or suggest color matching decisions in the design field, such as product design, clothing design, interior design, graphic design and web design.

10‧‧‧System

11‧‧‧ Word Segment Module

111‧‧‧Stutter Chinese word segmentation algorithm

13‧‧‧Expansion module

131‧‧‧ word embedding model

15‧‧‧ Semantic Correspondence Module

17‧‧‧ Color Construction Module

R‧‧‧ Narrative sentences

21‧‧‧Expanded vocabulary

23‧‧‧ color image vocabulary

25‧‧‧ color combination

S01~S08‧‧‧Step

FIG. 1 is a schematic diagram of the system architecture of the system corresponding to natural language to color combination of the present invention.

FIG. 2 is a schematic flowchart of steps of the method of the present invention corresponding to natural language to color combination.

The following specific examples are provided to illustrate the present invention, and those who are familiar with this field can undoubtedly understand the advantages and effects after reading the invention of this specification.

It should be understood that the structure described in this specification and the accompanying drawings, The proportions, sizes, etc. are disclosed only to match the content of this specification, so that those familiar with the field can easily understand and read, and are not intended to limit the present invention to specific situations and have no technical intent. Any modifications to the structure, changes in the proportional relationship, or adjustments in dimensions should be included in the scope of disclosure of this specification without affecting the manufacturability and achievement of this specification. Changes or adjustments in the relative relationship without substantially changing the technical content should also be deemed to fall within the scope of implementation.

Please refer to FIG. 1, the present invention provides a system corresponding to natural language to color combination, wherein the system 10 includes a word segmentation module 11, an expansion module 13, a semantic correspondence module 15, and a color construction module 17. It can be constructed by a computer with a processor, memory or storage device.

Users can input natural language Chinese narrative sentences, arbitrary vocabulary, or color image vocabulary through any interface. The word segmentation module 11 uses Jieba Chinese word segmentation algorithm 111 to distinguish the input content into multiple Vocabulary.

The expansion module 13 uses a word embedding model 131 to expand each of the plurality of vocabularies into a similar vocabulary group having a plurality of similar vocabularies, and aggregates the similar vocabulary groups of the plurality of vocabularies into a total vocabulary In extension set 21, the word embedding model 131 is constructed by a shallow and double-layer neural network trained by the word2vec algorithm.

The semantic correspondence module 15 calculates the vocabulary similarity between each similar word of the total vocabulary extension set 21 and the preset color image vocabulary 23 through the similarity equation, and sorts the vocabulary similarity based on the numerical value to select The vocabulary similarity ranks at least one color image vocabulary 23 with a high rank.

The color construction module 17 outputs the at least one color image vocabulary 23 corresponding to the preset color combination 25, wherein the color combination 25 of the color construction module 17 is a function library created by the drawing tool module Matplotlib, which It is the visual operation interface of the Python programming language and its numerical mathematics expansion package NumPy, which draws RGB values based on hue and hue arrangement and combination on the xy coordinate axis and displays the color blocks and corresponding values, and can be saved as a png image file.

After the conversion is calculated by the aforementioned system 10, any Chinese sentence or vocabulary entered by the user can be divided into multiple vocabularies by the word segmentation module 11, and then expanded by the expansion module 13 into similar vocabularies of similar word meanings Group, calculate the aggregated vocabulary expansion set 21 through the semantic correspondence module 15 to obtain the most similar multiple color image vocabulary 23, and then output the corresponding one of the multiple color image vocabulary 23 through the color construction module 17 Or multiple color combinations 25, which include corresponding RGB values, the aforementioned system 10 can automatically convert natural language into color combinations 25, which can assist or suggest color matching decisions in the design field, such as product design, clothing design, interior design , Graphic design and web design.

Referring to FIG. 2, the present invention provides a method corresponding to natural language to color combination, including:

In step S01, a narrative sentence R is input, which is a narrative sentence R under the natural language structure of the design requirement.

In step S02, the word segmentation module is used to distinguish the narration sentence R into multiple vocabularies, wherein the word segmentation module converts the narration sentence R into multiple vocabulary r ₁ through the Jieba Chinese word segmentation algorithm r ₂ ,...,r _i …, r _m .

In step S03, each of the plurality of vocabularies is expanded to a similar vocabulary group having a plurality of similar vocabularies through a word embedding model; wherein, the word embedding model is a shallow and double layer trained by the Word2vec algorithm The neural network is constructed by expanding the word segmented word r _i into n similar words that are closest to the word r _i to become a similar word group e(r _i )={e ₁ ,e ₂ ,..., e _n }.

，其中，E(R)為總詞彙擴展集，r_i為多個詞彙，藉由該word2vec演算法擴展為m組的該相近詞彙組e(r_i)，且共有p個相近詞彙α。 In step S04, the similar vocabularies that aggregate the plurality of vocabularies form a total vocabulary expansion set, where the total vocabulary expansion set is calculated by the following formula:

, Where E(R) is the total vocabulary expansion set, and r _i is a plurality of vocabulary, which is expanded into m groups of similar vocabulary groups e(r _i ) by the word2vec algorithm, and there are p similar vocabularies α .

，其中，c _k 為該色彩意象詞彙c _k ，該總詞彙擴展集中有l個該相近詞彙α _l ，l為從1至p，sim(c _k ,α _l )為c_k與α_l所計算出的詞彙相似度。 In Step S05, calculating the total of each of the similar word set and a preset vocabulary extension of Color Image vocabulary word similarity which in similarity obtained through the similarity-based equation S _k, S _k of the similarity equation is the following equation:

, Where c _{k is} the color image vocabulary c _k , there are l the similar vocabulary α _{l in} the total vocabulary expansion set, l is from 1 to p , and sim ( c _k ,α _l ) is calculated by c _k and α _l Vocabulary similarity.

In step S06, sort the vocabulary similarity based on the numerical value; and in step S07, select at least one color image vocabulary with a higher sorted vocabulary similarity. For example, the calculated vocabulary similarity is passed from high to low After sorting, the color image vocabulary c _x1 , c _x2 , c _x3 represented by the top three with the highest numerical value is used as the desired result.

In step S08, output at least one selected color combination corresponding to the preset color image vocabulary, wherein the specific embodiment of the present invention is that the color combination is based on the 11 hues (Hue) listed by Munsell , Such as red, orange, yellow, yellow-green, green, blue-green, blue, blue-violet, purple, magenta and achromatic, and then arranged in 12 different tones (Tone), such as sharp tone, emphasis, bright Tone, light tone, lightest tone, weak tone, weak tone, astringent tone, dull tone, dark tone, dark tone and darkest tone, a total of 130 colors are obtained, each color combination has 9 groups of RBG three Color combinations, including a total of 1170 RBG three-color combinations.

At least one of the selected color combinations corresponding to the color image vocabulary can be selected by using the function library of the drawing tool module Matplotlib. The visual operation interface of the Python programming language and numerical expansion package NumPy serves as the definition and the RGB value. Array operation and save it as a png image file to output the RBG three-color combination to which the color image vocabulary corresponds.

The color image vocabulary preset by the present invention and the corresponding color combination are based on the color image scale established by scholar Kobayashi Koshun, which illustrates the relationship between color and image, and then used a similar method to study The RBG three-color combination and its corresponding color image vocabulary are widely used in the color matching needs of daily life, especially in the design field.

In order to make the color image vocabulary easily convert into the corresponding color group To be specific, the present invention is based on the relationship between the book "Color Image Scale" by scholar Kobayashi's "Color Image Scale" for human perception and color, and the 180 color images and 39 expansion words in this book are unified. In order to become 222 color image vocabulary, and use word2vec as a natural language processing tool, any word similar to color image vocabulary is assembled, constructed, and trained to obtain a word embedding model 131.

The following describes an exemplary embodiment of the present invention:

In step S01, enter the narrative sentence R = "Don't care forever, only care about once owned".

In step S02, the word segmentation module is used to distinguish the narration sentence R into multiple vocabularies. The word segmentation module converts the narration sentence R into a plurality of vocabulary r through the Jieba Chinese word segmentation algorithm _i , such as "don't care", "forever", "only care", "once", and "own".

In step S03, each of the plurality of vocabularies is expanded to a similar vocabulary group through the aforementioned word embedding model, that is, each of the vocabularies r _{i is} expanded to the n closest to its own vocabulary r _i Similar vocabulary to become similar vocabulary group e(r _i )={e ₁ ,e ₂ ,…,e _n } as follows: For example, expand “don’t care” to “don’t care”, “don’t mind”, “don’t care”, “yes "," and "speak", "can't see", "have never seen", "do not understand", "do not count", and "do not understand" and other similar vocabulary group.

For example, expand ``Eternal Time'' into ``Jinzhiyuye'', ``Weiqing'', ``Qimeng'', ``Jinyumantang'', ``Weiyuan'', ``Jiangshan Beauty'', ``Like the Water'', ``Caiyunfei'', `` Similar vocabulary groups like "Happy Spring" and "Wen Zhaolun".

For example, expand "only care" into "love you love", "dare", "sharp" Similar phrases such as "to", "I Lai", "I love very much", "Jinlun", "I ask", "I control", "Please forgive", and "To be determined".

For example, expand "once" to "once", "return", "and once", "once", "once", "never", "never", "frequent", "many" , And "never" and other similar vocabulary groups.

For example, expand "own" to "have", "have", "have", "have", "have", "have", "have", "have", "have", and "show" Similar vocabulary.

In step S04, the similar vocabulary group of each of the plurality of vocabularies is calculated as a total vocabulary expansion set, where the total vocabulary expansion set is E(R)={"don't care", "do not mind", " It doesn't matter"... "hold", "have", "with", "show"}.

，其中，c _k 為該色彩意象詞彙c _k ，該總詞彙擴展集中有l個該相近詞彙α _l ，l為從1至p，sim(c _k ,α _l )為c_k與α_l所計算出的詞彙相似度，經過計算如下：S₁=sim(楚楚動人，E(R))=0.5658。 In step S05, the similarity equations are used to calculate the lexical similarity between each similar word of the total vocabulary expansion set and each preset color image vocabulary. The similarity equation S _{k is} as follows:

, Where c _{k is} the color image vocabulary c _k , there are l the similar vocabulary α _{l in} the total vocabulary expansion set, l is from 1 to p , and sim ( c _k , α _l ) is calculated by c _k and α _l The similarity of the vocabulary is calculated as follows: S ₁ =sim(chuchu, E(R))=0.5658.

S ₂ =sim(romantic, E(R))=0.5334.

S ₃ =sim(fresh, E(R))=0.5019.

...

S ₂₂₃ =sim(candid, E(R))=0.5159.

In step S06, the vocabulary similarity is sorted based on the numerical value, and the vocabulary similarity S ₁ , S ₂ , S ₃ , ..., S _{223 is} sorted as follows: S ₁₃₂ =sim(dashing, E(R))=0.6240.

S ₂₀ =sim(free and easy, E(R))=0.6144.

S ₁₃₁ =sim(acid, E(R))=0.6103.

...

In step S07, at least one color image vocabulary with a higher similarity degree of the vocabulary is selected, for example, after the calculated S _k similarity is sorted from high to low, the top three with the highest value are represented The color image vocabulary c ₁₃₂ , c ₂₀ , c ₁₃₁ , that is, "chic", "free and easy", "acid" output.

Step S08, output at least one selected color combination corresponding to the preset color image vocabulary. From the aforementioned 1170 RBG three-color combinations, the three corresponding to "chic", "free and easy" and "acid" are retrieved The RGB three-color combination corresponding to the color image vocabulary.

Through the above, the narrative sentences of the natural language input by the user, "don't care forever and only for once", through the method or system of the present invention, the three color image words of "chic", "free and easy" and "acid" are obtained In this way, the spoken narrative sentences of natural language of general users can be connected to the professional color image vocabulary used by designers through the method and system of the present invention to establish the relationship between the two and further let the design The designer or automated design calculation module can use this to carry out appropriate color planning to complete the subsequent design or color matching.

The above detailed description is a specific description of a feasible embodiment of the present invention, but this embodiment is not intended to limit the patent scope of the present invention, and any equivalent implementation or change without departing from the technical spirit of the present invention should be included in The patent scope of this case.

To sum up, this case is not only innovative in terms of technical ideas, but also can improve the above-mentioned multiple functions compared to conventional articles. It should be based on the legal requirements for patents for inventions that fully meet the novelty and progress. You must file an application according to law and urge your office to approve this document. Invention patent application, in order to encourage invention, to feel good.

S01~S08‧‧‧Step

Claims (10)

A method corresponding to natural language to color combination includes: dividing a narrative sentence into multiple words through a word segmentation module; expanding each of the multiple words into a close word with multiple close words through a word embedding model Group; the similar words that aggregate the multiple words form a total vocabulary expansion set, and the total vocabulary expansion set is calculated by the following equation:

, Where E(R) is the total vocabulary expansion set, r _{i is} the multiple vocabulary, which is expanded into m groups of similar vocabulary groups e ( r _i ) by the word2vec algorithm, and there are p similar vocabularies α ; calculating a total extension word set to the close similarity of each vocabulary word with a preset vocabulary of color images, through which the similarity-based computing equation S _k, S _k equation is based on the similarity

, Where c _{k is} the color image vocabulary, the total vocabulary expansion set has l the similar vocabulary α _l , l is from 1 to p , and sim ( c _k ,α _l ) is calculated by c _k and α _l Vocabulary similarity; sort the vocabulary similarity based on numerical value; select at least one color image vocabulary with a higher similarity of the vocabulary; and output at least one selected color combination corresponding to the preset color image vocabulary. According to the method described in item 1 of the patent application scope, wherein the word segmentation module divides the narrative sentence into multiple vocabularies through a Chinese word segmentation algorithm of Jieba. The method according to item 1 of the patent application scope, wherein the word embedding model is constructed by a shallow, two-layer neural network trained by the word2vec algorithm. The method according to item 3 of the patent application scope, wherein the color combination is an RGB color combination based on hue and hue arrangement combination. A system corresponding to natural language to color combination includes: a word segmentation module to divide the narrative sentence into multiple words; an expansion module to expand each of the multiple words through a word embedding model to Similar vocabulary groups with multiple similar vocabularies, and merging the similar vocabularies of the multiple vocabularies into a total vocabulary expansion set; a semantic correspondence module for calculating the respective vocabulary of the total vocabulary expansion set and the preset color The vocabulary similarity of the image vocabulary, and sort the vocabulary similarity based on the numerical value to select at least one of the color image vocabularies with the highest ranking of the vocabulary similarity; and a color construction module to output at least one selected color image The color combination corresponding to the vocabulary preset. According to the system described in item 5 of the patent application scope, the word segmentation module divides the narrative sentence into multiple vocabulary through the Chinese word segmentation algorithm of Jieba. The system according to item 5 of the patent application scope, wherein the word embedding model of the expansion module is constructed by a shallow and double-layer neural network trained by the word2vec algorithm. The system according to item 7 of the patent application scope, wherein the total vocabulary expansion set is calculated by the following equation:

, Where E(R) is the total vocabulary expansion set, and r _{i is} the multiple vocabularies, which are expanded into m groups of similar vocabulary groups e ( r _i ) by the Word2vec algorithm, and there are p similar vocabularies α . A system according to item 8 of the patent range, wherein the system module corresponding to semantic similarity equation S _k by calculating the total of each of the extended set of words similar to the words of the predetermined vocabulary word similarity Color Image , The similarity equation S _k system

, Where c _{k is} the color image vocabulary, the total vocabulary expansion set has l the similar vocabulary α _l , l is from 1 to p , and sim ( c _k ,α _l ) is calculated by c _k and α _l Vocabulary similarity. The system according to item 5 of the patent application scope, wherein the color combination of the color construction module is an RGB color based on hue and tone arrangement combination written in the Python programming language.

Images