A Study of the Fundamental Role of Color Perception in the Design of Graphic Composition

Color perception is an ability that everyone has, but it is only due to different living environments, the influence of personal life experiences, or the influence of traditional cultural factors that can cause differences in color perception [1–2]. The human eye is a complex and sophisticated visual organ, which can perceive and identify colorful colors. Color is determined by the wavelength and intensity of light, and color perception is a complex and sophisticated biological process, which is affected by a combination of physiological, psychological and cultural factors [3–4]. The conditions of color perception include brightness, color and direction of light, reflectivity, transparency and material of the object surface, and physiological and psychological state of the observer. Among them, the brightness and color of light play a decisive role in color perception, which is related to the individual’s comfort and perception [5–6]. Meanwhile, space, object form, geometric structure, and object material also affect the accuracy and sense of depth of color perception [7–10]. In addition, the physiological and psychological state of the observer will also have an impact on color perception, such as emotion, memory, and culture [11].

Color it not only conveys emotion and guides the eye, but also increases the expressiveness and attractiveness of art works [12–13]. By perceiving and controlling the saturation and brightness of colors, artists can create different atmospheres and moods, thus guiding viewers to different understandings and feelings about the work. Color can also be used to communicate product positioning and brand image, and through the choice and combination of colors, it can help products stand out in the market and attract users’ attention [14]. In advertising design, brand marketing and visual art creation, the use of color is related to the communication effect of the work and the audience’s experience. Through a deep understanding of color psychology and color perception, designers and artists can make better use of color to express their intentions, attract the audience, and achieve visual communication effects and marketing purposes [15–17]. In addition, color perception is the most sensitive factor in visual presentation. Color processing in graphic design occupies a very important position, the overall effect of the color needs to be eye-catching and individuality, can catch the consumer’s eye, can produce different feelings through the symbol of color, to achieve its purpose. In the visual expression of graphic design, the commonality and individuality of color have their own independent connotations, but also mutual illumination, combined with each other [18–19]. Weakening the personalized expression of color tone and emphasizing the sharpness of color is tantamount to weakening the market competitiveness of its products, such as yellow and orange [20]. It can be seen that color perception plays a crucial role in product design.

Image quality evaluation using color features can better simulate the human visual color perception process. Therefore, this paper proposes a color image quality evaluation algorithm model based on color visual features. The algorithm is based on the sequential learning theory, which transforms the quality prediction problem into a sequential problem and then weakens it into a preference learning problem. In this algorithm, three color features, namely, color harmony, color contrast and color richness, and related luminance features are extracted to quantify the distortion degree of color images, and combined with a machine learning classification algorithm to solve the preference learning problem. The method is applied to the design of graphic composition to judge the effect of color perception by six indexes, namely, the number of gaze points, the average gaze time, the total gaze time, the number of eye jumps, the average speed of eye jumps, and the average amplitude of eye jumps.

2

Color perception and graphic design

2.1

Color perception theory

Color perception theory is a branch of the discipline that studies how color affects human mental states and behavioral patterns, aiming to understand how humans perceive, interpret, and respond to color, with color psychology occupying a central position. Color psychology focuses on how color affects human emotions, cognition, and behavior [21], intending to reveal the profound connection between color and psychology.

Different colors can stimulate different emotional responses. Therefore, the emotional association of color is one of the basic contents studied in color psychology. For example, red is often associated with enthusiasm, energy and passion, while blue is usually associated with calmness, peace and trust. These emotional associations are formed based on cultural and personal experiences and therefore may vary across cultural backgrounds and social groups. For example, in Western cultures, white is often associated with weddings and celebrations, while in Eastern cultures it is commonly used for funerals and mourning occasions.

Color not only affects people’s emotions, but also their mental state and behavioral decisions. Studies have shown that color can change an individual’s state of mind in a short period of time, which in turn affects their purchasing behavior. For example, bright colors can attract attention and inspire impulse shopping. Soft colors, on the other hand, help to create a comfortable and relaxing atmosphere and are suitable for the design of spaces for long stays. Additionally, colors can affect people’s cognitive speed and productivity. For example, warm colors speed up the heart rate and increase alertness, making them suitable for work environments that require quick decision-making. Cooler colors, on the other hand, help calm thinking and are suitable for task scenarios that require concentration.

2.2

The role of color in graphic composition design

In graphic composition, color can also play a pivotal role in the expression, on the basis of producing distinctive visual effects, while creating a sensual atmosphere, such as bringing a surging visual impact, or creating a dim nostalgic scene, which is from the sensitivity of people to different colors to bring about different emotional expression of the charm of color, and greatly play the effect of graphic design. In the different use of hue, brightness and saturation, the image of emotional elaboration can make people appreciate the meaning of graphic design. Therefore, the expression of color can not be ignored in graphic composition.

Color can not only help designers better convey information, but also affect the audience’s perception, emotion and behavior. Designers can convey specific emotions and atmosphere by choosing appropriate colors, thus enhancing the attractiveness of the design. It can also help designers to convey messages more effectively. In addition, color can be used to emphasize important information or to differentiate between different design elements. It can also be used to create and maintain a brand image. By using colors that are consistent with the brand’s positioning and values, designers can help the brand create a unique image in the minds of consumers.

3

Color image quality evaluation method based on color perception characteristics

This paper designs a color image quality evaluation method based on color perception characteristics, and applies the technique to graphic composition design to enhance the visual appeal of graphic design.

3.1

Image color visual feature extraction

3.1.1

Color Harmony

In this paper, we design complex color harmony models to quantify the color harmony of color images with complex color distributions. 1)

Image local region acquisition

When the color distribution of an image is complex, the image can be regarded as the sum of local regions, and the color distribution of the image within the local regions is relatively simple. By calculating the color harmony of the local regions and fusing them together, the color harmony of the whole image can be obtained. In this paper, the image is divided into local blocks of the same size, which are combined by simple color patterns, and then the descriptors of the local blocks of the image are constructed, and the probability distribution of these descriptors in the whole image is calculated, which is used as the color harmony of the whole image.

2)

Constructing color descriptors

In this paper, the main color of the local region of the image block is used to represent the color harmony of the image by calculating the relative difference between the hue, saturation and luminance values of the local image block and the main color. Firstly, the local image block is converted from RGB color space to Munsell color space, the primary color of the image block is calculated based on the hue value, then the hue value of each pixel in the image block is subtracted from the primary color to find the absolute value, and finally their frequency distribution histograms are calculated. The obtained frequency histogram is used to describe the local image block.

3)

Color quantization and image representation

Through the above steps, a large number of image local descriptors are obtained. Then the obtained image descriptors are quantized using K-means mean clustering [22] to obtain image codes. Then the histogram of the frequency distribution of these codes over the whole image is computed, and the vector characterizing the image color harmony f_ch is obtained by concatenating the histograms.

3.1.2

Color contrast

In this section the color contrast of the image is calculated using Michelson’s contrast formulae (1), (2) and the luminance contrast formulae (3), (4), (5) from the structural similarity algorithm extended to the color data in the YIQ color space, and it is denoted as f_cc = [f₁, f₂, f₃, f₄, f₅]: 1 $f_{1} = \frac{1}{M} \sum_{i = 1}^{M} \ln | \frac{I_{i} - I_{m e a n}}{I_{i} + I_{m e a n}} |$ 2 $f_{2} = \frac{1}{M} \sum_{i = 1}^{M} \ln | \frac{Q_{i} - Q_{m e a n}}{Q_{i} + Q_{m e a n}} |$ 3 $f_{3} = \frac{σ_{I} (x) + a}{σ_{I}^{2} (x) + σ_{Q}^{2} (x) + a}$ 4 $f_{4} = \frac{σ_{Q} (x) + b}{σ_{I}^{2} (x) + σ_{Q}^{2} (x) + b}$ 5 $f_{4} = \frac{2 I (x) \cdot Q (x) + c}{σ_{I}^{2} (x) + σ_{Q}^{2} (x) + c}$ where I, Q represents the color channels in the YIQ color space, respectively, covering all color data of the color image. M represents the number of pixels and a, b, c is a constant.

3.1.3

Color richness

The color richness of an image is classified into different levels, ranging from extremely brilliant to colorless, based on the richness of the image’s color. By studying the distribution of image pixels in the CIELAB color space, the color richness f_cf is defined as follows: 6 $f_{c f} = ω_{1} σ_{a} + ω_{2} σ_{b} + ω_{3} σ_{a b}$ 7 $σ_{a b} = \sqrt{σ_{a} + σ_{b}}$ where σ_a, σ_b represent the standard deviation of the a and b axes, respectively. σ_ab represents the standard deviation of the triangular lengths in ab-space.

3.2

Quality evaluation model training based on ranking learning

The images in the database are viewed by observers in a specific test environment to be tested, scored manually according to their own visual perception, and finally the average of the scores given by all the observers is calculated to obtain an average subjective score MOS value, which is used as a result of the quality evaluation of the image [23]. Using the difference between the MOS values of the two images in the database, it is possible to determine whose quality of these two images is superior. The image pair that can judge the difference in quality between two images is called “Preferred Image Pair” (PIP). The training of the quality evaluation model based on ranking learning first requires the construction of a PIP image pair, and then compares the quality of the two images to design a preference class label (PL) for the image pair. For example, if the subjective quality score of image l₁ is MOS₁, and the subjective quality score of image l₂ is MOS₂, we use I₁ and I₂ to construct a preference image pair (I₁, I₂). We design a preference label by comparing the values of MOS₁ and MOS₂. If the quality of l₁ is better than the quality of l₂, i.e., MOS₁ – MOS₂ > 0, then the preference label y₁ is equal to 1, and vice versa, y₁ is equal to -1. At this time, we obtain a pair of PIP image pairs and its preference class label.

In the training phase, a training dataset is created by randomly selecting N pair of PIP images on a database in the manner described above and setting preference classes on them. Assuming that a database has N images, i.e., I = {I_I, I₂ ⋯ N_N}, and their corresponding image MOS values can be represented as MOS = {MOS₁, MOS₂ ⋯ MOS_N}, the PIP image pair P constructed on that database can be represented as follows: 8 $P \subseteq {(I_{i}, I_{j}) | y_{i} = s i g n (M O S_{i} - M O S_{j}), i, j = 1, \dots N}$ where (I_i, I_j) represents the selected image pairs and y_i represents the preference class labels. After creating the training dataset, the color visual features f_ch, f_cc, f_c of the color images and the luminance features in BLIINDS2 are extracted according to the method in Section 3.1. The features of two images in an image pair are subtracted and the resulting feature difference vector is utilized to represent a preference image pair. At this point, the problem of mapping the feature difference vectors of the images and the preference class labels is transformed into a binary classification problem, where a classifier is learned using a support vector machine (SVM). When the feature vector difference x_best of the test image is fed into the learned classification algorithm, the corresponding preference class label y_best is obtained.

3.3

Image Quality Evaluation Measure

In the testing stage, the test image I_best and other images are composed into different PIP images and the preference class label corresponding to the feature vector difference x_best1 ⋯ x_bestN of the PIP image pairs of the test image is obtained by the trained classifier y_best1 ⋯ y_bestN. Then the preference class label is summed to obtain the gain g_text of the test image I_best as shown in Eqn. (9) and since the quality of the color image is directly proportional to its gain, the evaluation of the quality of the color image is evaluated using the gain g_best measure Q is expressed as equation (10): 9 $g_{b e s t} = \sum_{b e s t i = 1}^{N} y_{b e s t i}$ 10 $Q = a g_{t e s t} + b$

Setting the interval of the perceived quality score of the image to [0,100], and based on the fact that the image with maximum gain (n – 1) corresponds to the optimal quality score and the image with minimum gain –(n – 1) corresponds to the worst quality score, a system of linear equations (11) can be obtained, and by solving the equations the values of a and b can be obtained: 11 $[\begin{matrix} n - 1 & 1 \\ - (n - 1) & 1 \end{matrix}] [\begin{array}{l} a \\ b \end{array}] = [\begin{matrix} 100 \\ 0 \end{matrix}]$

4

Experimental results and analysis

4.1

Validation of color image quality evaluation methods

4.1.1

Consistency experiments

The database chosen for this experiment is TID2013, which is an upgraded and improved version of TID2008. This database contains 3000 distorted images with 24 different distortion types, each type is divided into 5 different distortion levels. This database has 8 types of color related distortion, and some mixed distortion types. The richness of the image content, the variety of distortion types, and the large number of images make it more reasonable to validate the image quality evaluation algorithm using TID2013.

We divided the database into mutually non-overlapping training and test sets, performed 100 program operations within each set, and averaged the obtained Pearson’s linear correlation coefficient (PLCC) and Spearman’s rank correlation coefficient (SROCC) values to compute as the evaluation index of this algorithm’s consistency with subjectivity. In the experiment, 80% of the images were randomly selected to form the training set and the remaining 20% were used as test images. That is, 20 sets of original images corresponding to the distortion are done for training and the remaining 5 sets are done for testing. Images Since this algorithm does not use any information of the original images, the comparison methods in this experiment are chosen to be non-referential type methods, which include: BIQI, BLIINDS2, DIIVINE, CORNIA, NIQE, QAC, IL-NIQE, and SRNSS.The results of the subjective and objective consistency comparison between this paper’s method and the above algorithms are shown in Fig. 1. The PLCC and SROCC values of this paper’s method are 0.667 and 0.563, respectively, which achieve a more satisfactory subjective and objective consistency. Compared with DIIVINE, this paper’s method does not select a large number of features and does not require a large amount of computing power to learn the mapping, and achieves a better overall result with a simple and small-scale computing power.

4.1.2

Individual distortion type experiments

In addition to the overall performance, we verified the effectiveness of this method for each distortion type. To be consistent with the way the original training model for the comparison experiment was done. In this experiment, we use the distorted images from the LIVEdatabase II database as the training set and then do the test on TID2013. A comparison of the results of the single distortion experiment is shown in Figure 2. The method of this paper can effectively perform quality evaluation for many kinds of single distortions, hybrid distortions, and color distortions, especially for the four kinds of distortions related to color changes, namely #10, #16, #17, and #23, the quality evaluation results achieved by the method of this paper are 0.9163, 0.2356, 0.5343, and 0.8864, respectively.All the comparative methods’ results on these kinds of distortion types are lower.

4.2

Graphic Composition Design Eye Movement Evaluation Analysis

The study applies the color image quality evaluation method based on color perception characteristics designed above to graphic composition design, and designs eye-movement evaluation experiments to judge the visual attractiveness of the design works under this method.

4.2.1

Graphic design eye movement indicators

This chapter discusses the relationship between various eye movement indicators and the visual attractiveness of design works, and selects a number of eye movement statistical analysis indicators (including the number of gaze points, the average gaze time, the total gaze time, the number of eye jumps, the average speed of eye jumps, the average amplitude of eye jumps, a total of six indicators), and analyzes the difference in eye movement between graphic designs based on the design of color perception and ordinary works. The basic meanings of each index are shown below:

Gaze count (s): the total number of gaze points within an area of interest.

Average gaze time (s): the average gaze duration of each gaze point.

Total gaze time (s): the sum of the gaze duration of all gaze points within a region of interest.

Number of eye jumps (s): the total number of eye jumps within a region of interest.

Mean eye-roll speed (deg/s): the mean speed of each eye-roll.

Mean eye-roll amplitude (degrees): the mean amplitude of each eye-roll.

4.2.2

Analysis of eye movement metrics of works under different methods

The eye movement data of graphic design works (1~9) and general graphic design works (10~18) under the method of this paper were imported into EXCEL for data processing, and the average value of each index was taken.

Comparing the number of gaze and the number of eye jumps found that the statistical results of the two data are shown in Table 1. The higher number of gaze counts and eye hopping counts represent more attention, indicating that some factors of these pictures or for some reasons attracted the attention of the subjects. The range of the mean values of the number of gaze and the number of eye jumps of the works designed under the method of this paper are between 19.379~23.51 and 12.158~15.89, respectively, which are higher than the mean values of the number of gaze and the number of eye jumps of the ordinary graphic design works.

Table 1.

The number of eyes and the number of eye beats

Photo number	Number of eyes/time	Jump frequency/time	Photo number	Number of eyes/time	Jump frequency/time
1	20.5	13.085	10	13.603	10.008
2	23.002	13.063	11	12.554	8.149
3	19.379	13.29	12	11.058	9.314
4	19.423	13.84	13	13.75	8.904
5	22.655	14.016	14	15.018	10.819
6	23.51	13.992	15	11.77	10.522
7	19.725	12.158	16	12.873	9.705
8	20.728	15.89	17	13.113	8.891
9	22.372	14.563	18	14.334	10.885

The mean eye-beat velocity is equal to the mean eye-beat amplitude divided by the eye-beat duration, which changes with different eye-movement behaviors, and in this experiment, the eye-beat duration is not in the category of eye-movement data to be studied, so we do not make too much speculation. However, the positive correlation between mean eye-beat speed and mean eye-beat amplitude is reflected in this experiment. Eye hop amplitude refers to the distance between the two gaze points before and after the occurrence of an eye hop, and the larger the mean eye hop amplitude, the more distinctive the features are in these pictures and easier to search. The statistical results of the mean eye-jump speed and mean eye-jump amplitude are shown in Table 2. The average eye-jump speed and average eye-jump amplitude of work 9 are 107.344°/s and 4.991°, respectively, which are the most distinctive features among the works designed based on the method of this paper. While the highest mean eye-jump velocity and mean eye-jump amplitude of ordinary works were 99.664°/s and 3.771°, respectively, with insignificant features.

Table 2.

The average eye jump velocity and the average eye jump amplitude

Photo number	Average eye speed(°/s)	Average eye jump amplitude/°	Photo number	Average eye speed(°/s)	Average eye jump amplitude/°
1	97.401	4.086	10	88.962	3.259
2	98.947	4.18	11	91.814	3.353
3	99.045	4.275	12	92.017	3.396
4	102.206	4.365	13	92.59	3.436
5	102.647	4.481	14	95.192	3.467
6	103.114	4.602	15	96.017	3.595
7	103.608	4.914	16	97.396	3.596
8	105.007	4.982	17	98.017	3.62
9	107.344	4.991	18	99.664	3.771

The total gaze time and average gaze time statistics for the different pictures are shown in Table 3. The higher the total gaze time, it means that the aesthetics of the picture or other reasons attracted the attention of the subjects, and the subjects spent a lot of time on appreciating or understanding this picture. Mean gaze time is the average length of gaze for each gaze point. The total gaze time for pictures 1~9 ranged from 7.274s to 7.914s. This indicates that the subjects spent more time in the observation process compared to pictures 10~18.

Table 3.

The total fixation time of different photos is measured by time

Photo number	Total fixation time/s	Average time/s	Photo number	Total fixation time/s	Average time/s
1	7.274	0.303	10	6.029	0.202
2	7.422	0.306	11	6.139	0.206
3	7.469	0.306	12	6.242	0.225
4	7.569	0.323	13	6.395	0.227
5	7.723	0.356	14	6.492	0.228
6	7.761	0.363	15	6.509	0.238
7	7.78	0.367	16	6.59	0.25
8	7.911	0.37	17	6.893	0.295
9	7.914	0.385	18	6.91	0.296

5

Conclusion

The study designed a color image quality evaluation model based on sequential learning for enhancing the application of color perception in graphic composition design. The PLCC and SROCC values of the method are 0.667 and 0.563, respectively, achieving a relatively satisfactory subjective and objective consistency. It shows that using color features to construct a color image quality evaluation algorithm is a way to evaluate the quality of color images. In the eye movement evaluation experiment of the graphic works designed by applying the method of this paper, the average values of six indexes, namely, the number of gaze points, the average gaze time, the total gaze time, the number of eye jumps, the average speed of eye jumps, and the average amplitude of eye jumps, were higher than those of ordinary works. It shows that the application of higher quality color perception can enhance the visual attractiveness of graphic composition design.

Lingua:: Inglese

Frequenza di pubblicazione:: 1 volte all'anno
Argomenti della rivista:: Scienze biologiche, Scienze della vita, altro, Matematica, Matematica applicata, Matematica generale, Fisica, Fisica, altro

Feed RSS della rivista

A Study of the Fundamental Role of Color Perception in the Design of Graphic Composition

Bing Li

Shuo Wang

Pubblicato online: 19 mar 2025

Ricevuto: 29 ott 2024

Accettato: 10 feb 2025

DOI: https://doi.org/10.2478/amns-2025-0506

Parole chiaveColor perception, Machine learning, Eye movement indicators, Graphic composition design

© 2025 Bing Li et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Parole chiave
Color perception, Machine learning, Eye movement indicators, Graphic composition design