CN106354838A - Data visualization method based on semantic resonance colors - Google Patents

Abstract

The invention relates to a data visualization method based on semantic resonance colors. The method comprises the following steps of 1, setting a set of nouns of articles with colors to be distributed as a word set and obtaining a picture set related to the word set according to keywords of the nouns; 2, selecting a proper color palette, wherein the color palette comprises all colors of the articles with colors to be distributed; 3, generating a color histogram of pictures in the picture set, and according to the distribution phenomenon of the colors in the color histogram, obtaining the occurrence probability of the colors in the color palette in the color histogram; 4, according to the distinguishing degree of the colors in the color palette and the occurrence probability in the color histogram, distributing the colors in the color palette to the articles with the colors to be distributed in the word set, and forming and displaying keyword-color pairs. Compared with the prior art, the data visualization method has the advantages of reserving main colors, being accurate in color matching and the like.

Description

A kind of data visualization method based on semantic sympathetic response color

Technical field

The present invention relates to image processing field, especially relate to a kind of data visualization side based on semantic sympathetic response color Method.

Background technology

During data visualization, color method will play a significant role, and good color rendering intent will greatly be lifted Visual effect, for example, we can show the data of " ocean " with blue, or with peach come for keyword " love " colours, and different colors can help us to identify and distinguish between different classifications.

It is very easy to find, color is consistent with the meaning of word to accelerate cognitive process, otherwise then can hinder cognition, this interesting showing As referred to as " Stroop effect " (stroop effect), in fact, not being singly the title of color, people are accustomed to already by color Bind together with each conception of species, these natural and semantic congruence colors are referred to as " semantic sympathetic response color " (semantically resonant color), if energy effectively utilizes semanteme sympathetic response color, we can improve visual cognition Efficiency.

But the form of expression of existing word is excessively dull, black font in the whole text for word semantics recognition very Unfavorable, nor lift the reading interest of reader, the textual representation method of existing semanteme sympathetic response color is generally using related in a large number The background of black and white, as training set, when obtaining the rejecting of the background before color histogram, is often weeded out by picture, But because the corresponding body color of some vocabulary is exactly black and white, the color of itself is rejected by therefore this kind of method, lead to unite Meter is inaccurate.

Content of the invention

The purpose of the present invention be exactly provide to overcome the defect that above-mentioned prior art exists a kind of retain body color, The color matching accurately data visualization method based on semantic sympathetic response color.

The purpose of the present invention can be achieved through the following technical solutions:

A kind of data visualization method based on semantic sympathetic response color, comprises the following steps:

1) as word finder, the key word according to noun obtains associated therewith the article noun setting one group of color to be allocated Pictures；

2) choose suitable palette, this palette includes all colours of the article of color to be allocated；

3) produce the color histogram of every width picture in pictures, according to the distribution situation of color histogram in figure color, obtain Take the probability that the color in palette occurs in color histogram in figure；

4) according to the identification between the color in palette and the probability in the appearance of color histogram in figure, by palette The article of color to be allocated in word finder for the color assignment on, formed key word-color to and show.

Described step 2) in, described palette adopts 20 tone colour tables.

Described step 3) specifically include following steps:

31) by the way of pixel filtration, background rejecting is carried out to the picture in pictures；

32) pixel count is adopted to count the quantity that in picture, each color occurs；

33) colour type quantity being in front three as many classes, remaining colour type as few class, accordingly to every width Picture gives color label,

34) abandon the corresponding picture of few class, and according to the corresponding picture of many classes, black and white are added to color histogram In the probability statistics colour type of figure and palette, produce corresponding color histogram.

The method is further comprising the steps of:

5) vocabulary semantic with color that this method produces is made comparisons with artificial color assignment statistics, obtain color matching accurate Exactness.

Described step 32) in, in lab space to picture in the pixel count of each color carry out quantity statistics.

Compared with prior art, the invention has the advantages that

First, retain body color: background rejecting is carried out using pixel statistics, existing eliminates substantial amounts of effective letter Breath, filtering background is changed to pixel and filters by the present invention, does not have white and black, because both colors are seldom used in former palette In the displaying of data visualization, therefore filtered again in statistical pixel color histogram, carried out for single pixel.

2nd, match colors accurately: the probability statistics of version cooperation color histogram of being matched colors using 20 colors, so that the method for the present invention is matched colors Accurately.

Brief description

Fig. 1 is the pixels statisticses figure in lab space.

Specific embodiment

The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.

Embodiment:

1. obtain data

It is that the vocabulary such as vegetable to one group of particular category (" Fructus Lycopersici esculenti ", " Rhizoma Solani tuber osi ", " Capsicum annuum L. " etc.) carries out Google's picture searching, And the picture searched for is taken together, the color of all pixels counting all pictures is to produce color histogram.

For a key word, such as " Fructus Mali pumilae ", in addition to direct search " Fructus Mali pumilae ", this method can also be by additional Word obtaining more rich image data, such as " Fructus Mali pumilae+material ", so each key word will be obtained with two color histograms Figure.

2. candidate's palette

Be not arbitrarily take average be one color of a key word arrangement be exactly suitable.In visualization tasks In it is necessary to realize in all of classification a kind of take into account discrimination and semantic color arrange just meaningful.In order to obtain Significant color arrangement, algorithm employs the 20 tone colour tables of well-known visualization company tableau.In each visualization tasks Middle reasonably will be distributed according to semanteme for classification based on 20 kinds of colors in this palette.

3. color arrangement

After obtaining all of color histogram, rectangular histogram is made with Density Estimator to simulate the distribution of color, and according to it Calculate the probability that in candidate's palette, each color occurs, then take into account the identification between color, obtain every a pair " color-word " sympathetic response fraction " (affinity score) that remittance " is combined.These sympathetic response fractions feature between concept and color the strong of connection Degree.According to all sympathetic response fractions, candidate color is assigned to one by one in word finder so that total sympathetic response fraction using Hungarian method Highest, that is, Color Semantic is the strongest.

The core formula of algorithm be presented herein below:

$p\left(c\right|v,t)\&proportional;\underset{b\&element;t}{\σ}t\left(b\right)\exp \left(\frac{dist(b,c)}{\mathrm{\σ}}\right)$

$h\left(c\right|v,t)=-\underset{c\&element;c}{\σ}p\left(c\right|v,t)lnp\left(c\right|v,t)$

$p\left(c\right|v,t)\&proportional;max\left(sat\right(c),t)\·\[\frac{w_{c}p\left(c\right|v,t_c)}{h\left(c\right|v,t)}+\frac{(1-w_c)p\left(c\right|v,t_c)}{h\left(c\right|v,t)}\]$

$affinity(c,v)=\frac{p\left(c\right|v)}{h\left(c\right|v)}$

Finally, using the mass-rent resource on amazon mechanical turk it is desirable to participant is to identical word finder Mated with color set, and adjusted the parameter of automation algorithm accordingly, made the result that automatization distributes as close possible to artificial The scheme selecting, a, t two row are respectively algorithm and the artificial color allocation scheme producing it is seen that both have much approximate ground Side.

4. arithmetic result and experiment

In the experimental stage, compare the impact to rectangular histogram reading efficiency for three kinds of different color allocation, respectively expert The scheme that selection, algorithm generate and be randomly assigned.Wherein select expert scheme by business visual software tableau design Person is given, the upper limit as other scheme works and criterion.In an experiment, tested it will be appreciated that three kinds of rectangular histograms, and want rooting According to the upper information answers problem of figure, its response time is recorded and is used as the tolerance of rectangular histogram effect.And the result tested shows, For coloring height, there is the vocabulary being relatively fixed color, such as " sky ", " Fructus Fragariae Ananssae ", " Rhizoma Solani tuber osi " etc., have semantic sympathetic response Color allocation can improve about 10% response time, wherein the scheme of select expert slightly be better than algorithm produce scheme.But it is right In abstract, vocabulary that is lacking fixing color, such as " coca cola ", " Google ", " healthy " etc., semantic sympathetic response can only provide relatively Little improvement.

Three. case study

Because color arrangement is highly dependent on the image data of acquisition, the quality of therefore result is heavily dependent on former The process of data prediction, and statistical method afterwards is not provided that too many denoising correction.

See one group of test result, See Figure first.

Can see that the color arrangement of first three keyword is unsatisfactory, find that there are the following problems:

1. the picture of mass efficient is left out during the background of former algorithm is rejected.(former algorithm passes through image edge 's

Pixels statisticses, count and black, number of pixels in defined threshold for the white, if it exceeds whole edge 75%, then it is assumed that this picture is difficult to extract crucial colors information, is deleted.) this directly results in mass efficient information quilt Ignore, totally die down, and the weight of noise data is bigger.For example in " milk ", blue background is retained to be become Big interference, and effective picture of white background is ignored.

2. the feature of the former data maximum of " Bulbus Lilii ", " Paeonia suffruticosa " is that color is numerous and diverse, and key feature data is difficult to extract.This The result directly contributing during statistics is that the conjugation of single key word and candidate's color is average, and last result dispersion does not have concentration Property.

Four. improvement

First, in data acquisition phase, this method is using up-to-date baidu image more stable at home Search api, repairs the support to Chinese character set.

Secondly, modification background rejects algorithm.In view of eliminating substantial amounts of effective information before, therefore filtering background is changed to Pixel filters.Originally there is no white and black, because both colors are rarely used in data visualization in the palette of existing use The displaying changed.Therefore filtered again in statistical pixel color histogram, carried out for single pixel.

3rd, as shown in figure 1, will be once polymerized before statistics.For lab space, it is pressed l, a, b are positive and negative to be amounted to 6 directions are divided into 6 classes.The all pixels of single width figure are sorted out by color, is finally counted pixel count of all categories.Take at most Pixel count that class be this width figure label.

Once counted again after all of figure all carries color classification label, take all kinds of sum classifications of first three to be crowd Class, remaining three classes are few class, and the picture of these classifications is abandoned.

Finally, displaying to be improved.Needing to carry out expressive enough displaying to result under existence conditionses could be to knot Fruit is estimated.Can not possibly look to holding the coding of several colors as final output result.Can be realized from d3 storehouse The digital independent of front end and display.

Claims (5)

1. a kind of data visualization method based on semantic sympathetic response color is it is characterised in that comprise the following steps:

1) as word finder, the key word according to noun obtains relative figure to the article noun setting one group of color to be allocated Piece collection；

2) choose suitable palette, this palette includes all colours of the article of color to be allocated；

3) produce the color histogram of every width picture in pictures, according to the distribution situation of color histogram in figure color, obtain and adjust The probability that color in colour table occurs in color histogram in figure；

4) according to the identification between the color in palette and the probability in the appearance of color histogram in figure, by the face in palette Color is assigned on the article of the color to be allocated in word finder, formed key word-color to and show.

2. a kind of data visualization method based on semantic sympathetic response color according to claim 1 is it is characterised in that described Step 2) in, described palette adopts 20 tone colour tables.

3. a kind of data visualization method based on semantic sympathetic response color according to claim 1 is it is characterised in that described Step 3) specifically include following steps:

31) by the way of pixel filtration, background rejecting is carried out to the picture in pictures；

32) pixel count is adopted to count the quantity that in picture, each color occurs；

33) colour type quantity being in front three as many classes, remaining colour type as few class, accordingly to every width picture Give color label,

34) abandon the corresponding picture of few class, and according to the corresponding picture of many classes, black is added to color histogram with white In probability statistics colour type and palette, produce corresponding color histogram.

4. a kind of data visualization method based on semantic sympathetic response color according to claim 1 is it is characterised in that the party Method is further comprising the steps of:

5) vocabulary semantic with color that this method produces is made comparisons with artificial color assignment statistics, obtain color matching accurately Degree.

5. a kind of data visualization method based on semantic sympathetic response color according to claim 1 is it is characterised in that described Step 32) in, in lab space to picture in the pixel count of each color carry out quantity statistics.