Research on Creativity Generation and Innovative Expression Strategies of Graphic Design Based on Graphic Processing Technology

The development of logo design in the era of smart media is related to many aspects such as social economy, culture, media, technology, etc. Changes in the external environment present new requirements for logo design, leading to the development and iteration of logo designs.

The development of mass media also plays an important role in the development of logo design. People’s exposure to various information provides an opportunity for symbol consumption. The existence and development of mass intelligent media provide an excellent communication carrier for logo design, and different communication subjects realize the importance of logo design, which also makes it necessary for logo design to adapt to different media.

The development of science and technology in the era of smart media enables logo design to be presented in the form of dynamization or virtualization, and artificial intelligence, virtual reality and other technologies have deepened the influence of technology on various industries, which also prompts designers to incorporate more scientific and technological elements to strengthen the scientific and technological expression of logo design.

The business purpose of a brand is to obtain economic benefits, and a series of economic activities are also centered on this purpose. In the context of increasingly fierce market competition in the age of smart media, brand logo design can, to a certain extent, enhance the symbolic value of the brand.

From the visual level, most brand logos have increased color brightness, which is related to the sharpness of screen display adapted to smart devices. The compatibility of different media, more and more display channels, and the presentation of different media forms have certain differences, how to maintain consistency in the visual presentation, while enhancing the expressive power is an important demand for the logo design within the brand nowadays.

Most of the current image super-resolution models follow the same type of module design for the most basic residual network module in the network, which always performs feature mining at the same resolution in the feature space. The information in the feature space varies depending on the resolution, but there has been no distinction between the frequency components of the feature space in previous research on image super-resolution tasks.

Taking this as a starting point, this paper designs a new Multifrequency Fusion Attention Module (MFAB) for feature enhancement based on different frequency components in the feature space. The module is based on the assumption of a feature scale space and differentiates the application of information from different frequencies in the feature space. The attention mechanism assigns the extracted feature information of different frequencies to the original features. So that the information of different frequencies in the original feature space can have different weights and improve the efficiency of the corresponding frequency components in utilizing the features. On this basis, the lightweight design of the module reduces the burden of parameter count and computation brought to the network by module replacement.

In this section, the single-image super-resolution enhanced deep residual network (EDSR) network is used as the basic framework, and the new multi-frequency fusion attention network (MFAN) is obtained by replacing the proposed MFAB with the base residual module in EDSR.

The overall structure of the EDSR model is divided into three network modules: feature extraction, nonlinear mapping, and reconstruction.The EDSR is trained using the mean absolute error (MAE, also known as LI) loss function, defined in equation (8): (8)MAE(X,Y;θ)=∑i,jh,w|Y(i,j)−X(i,j)|h×w

The network takes low-resolution images as input to the network. After the feature extraction module it is fed into the feature enhancement module, which consists of several multi-frequency fusion attention modules (MFAB). The strengthened features are fed into the final image reconstruction module to obtain the final resolution-enhanced high-definition image as the output of the network. The most important part of the whole network model framework is the feature reinforcement module, which determines the efficiency of the whole super-resolution network to utilize the information in the feature space.

The module proposed in this section starts from a new perspective and dives deeper into the feature information corresponding to different frequencies in the feature space, based on the space theory of scales. The most basic residual module in the feature enhancement module has been improved, and the multi-frequency fusion attention module (MFAB) has been proposed.

In MFAB, the standard residual module is still used as the feature extractor, but a new module (MFF) that fuses multi-frequency information is added after this module, and MFAB is the product of the combination of the two.

From the structure of the Multifrequency Fusion Attention Block (MFAB), it can be seen that MFAB adds a new branch of attention to the traditional residual module. This branch focuses on feature mining in space with different resolutions, fully considering the importance of different frequency information of features. And then the information sampled in different frequency spaces is upsampled to the original resolution space for fusion, so as to obtain a weight map with different attention to different information.

It is not a bad idea to set the input feature of the ist MFAB to be f_i−1. The input feature is passed through the plain convolution module to obtain feature x_i, i.e: (9)xi=Conv3×3(ReLU(Conv3×3(fi−1)))

The feature resolution is reduced by three downsampling modules to get the new feature x_i2, x_i4, x_i8 at different resolutions, i.e: (10)xi2 = Downsample(xi,2) xi4 = Downsample(xi,4) xi8 = Downsample(xi,8)

And then the new features at each resolution go through Basic Block for feature extraction and information filtering to obtain the feature weight sampling s_i2, s_i4, s_i8 at each resolution, i.e.: (11)si2 = Conν1×1(ReLU(Conν3×3(xi2))) si4 = Conν1×1(ReLU(Conν3×3(xi4))) Si5 = Conν1×1(ReLU(Conν3×3(xi8)))

Sampling each of the obtained feature weights and then performing an upsampling operation to recover the resolution, the fusion yields the feature weights w_i in the original space, i.e: (12)wi=Conv3×3(Upsample(si2,2)+Upsample(si4,4)+Upsample(si8,8))

Weighting the original features after normalizing the weight map, × represents the Hadamard product of matrices, which together with the input features f_i−1 in the constant mapping branch gives the final output features f_i. i.e.: (13)fi=fi−1+xi×Sigmoid(wi)

In order to be easily embedded into existing image super-resolution networks to replace the existing residual module, this section lightens the algorithm after the Conv-ReLU-Conv output of the feature maps in the traditional residual branch, and before adding them to the constant mapping branch.

The scheme used in this section is to reduce the number of channels of the feature map and maximize the downsampling multiplicity, which corresponds to the additional modules in the network design and the corresponding 2-, 4-, and 8-fold downsampling multiplicities.

Given the number of input channels is C_in, the number of output channels is C_out, the input feature map size is (H_in, W_in), the output feature map size is (H_out, W_out), the convolution kernel size is K, the step size is 1, and the bias term is not taken into account, the standard convolution operation operates as: (14)FLOPS=(Cin×K×K+Cin−1)×Hout×Wout×Cout

With a channel reduction multiplicity of n, the computation of the standard convolution inside the lightweight multi-frequency fusion attention module proposed in this section will be reduced to: (15)FLOPS′=(Cinn×K×K+Cinn−1)×Hout×Wout×Coutn

Obviously FLOPS ≈ FLOPS′ × n², the computation of the attention branch in the new multifrequency fusion becomes 1n2 of the original one. Greatly reduces the proportion of the new multifrequency fusion attention branch, enabling the multifrequency fusion attention module proposed in this section to accomplish the replacement of the residual module in the ordinary super-resolution task without increasing the computation too much.

The image super resolution technique proposed in this paper is applied to dynamic graphic design.

Graphic design dynamization has the following innovations. Dynamization of graphic design is different from the previous video, animation, it can intercept the video highlights and add other elements of the design to meet the audience’s curiosity about the unplayed film, but also the static continuous animation to the dynamic form of presentation. So dynamic graphic design to short, new, strange, content production requirements to convey a sense of dynamism, the use of three-dimensional characters and other constituent elements constitute a dynamic model, because of its own specific two-dimensional space characteristics, the design of the work can not go beyond the spatial boundaries, so the designers are as much as possible in the content and presentation methods to expand the expression of the space, so that the dynamic and static phase balance.

To ensure the validity of the survey results, a total of 20 experts and scholars related to motion graphic design were invited to conduct the test. Among them, there are 4 teachers of visual communication design, 3 graduate students of art design, 3 doctoral students of design, 3 graduate students of digital media design, 4 dynamic designers, and 3 designers in the direction of user perception.

The dynamic graphic perceptual fluency factors, dynamic graphic cognitive and aesthetic themes were refined to constitute the specific components of dynamic graphic design evaluation based on perceptual fluency. Perceptual attributes, perceptual experience attributes, cognitive attributes, and aesthetic attributes are subdivided separately. Specifically, clear visual organization A1, coherent graphic movement A2, smooth camera movement A3, reasonable temporal order A4 and effective audio-visual integration A5 are categorized as perceptual attributes. Content familiarity B1, stylistic difference B2 and perceptual imagination B3 were categorized as perceptual experience attributes. Information readability C1, attention attraction C2, and memory retention C3 were categorized as cognitive attributes. Aesthetic pleasure D1, aesthetic expectation D2, and aesthetic interest D3 were categorized as aesthetic attributes.

The specific elements of the dynamic graphic design evaluation model based on perceptual fluency are as follows: 1)

Goal layer. The only element of the goal layer is the optimal design solution for dynamic graphics based on perceptual fluency.

Each tester was tasked with comparing the relative importance of each element between the same levels two by two. The scores were based on a scale of 1-9, and the geometric mean algorithm was applied to construct a judgment matrix and count the weight values of each design element. Then the consistency test was performed on the judgment matrix to ensure the logical correctness of the matrix. After the operation, the CR values are less than 0.1, and the consistency test is passed. The results of the consistency test are shown in Table 3.

In order to calculate the comprehensive weight ranking of each design element of the sub-criteria layer, the design elements of the sub-criteria layer are multiplied with the design elements of the corresponding criterion layer to obtain the comprehensive weights of the 14 design elements of the dynamic graphic, which are ranked according to the magnitude of the weight values.

The perceptual fluency-based weight prioritization of dynamic graphic design elements is shown in Figure 1. The first three design elements are clear visual organization A1=0.2012, aesthetic interest D3=0.1896, and coherent graphic motion A2=0.1287. Through the results of weight calculation, the four design attributes of the guideline layer are sequentially ranked as perceptual attributes, aesthetic attributes, cognitive attributes and perceptual experience attributes. It can be seen that in dynamic graphic design, the most important index for improving perceptual fluency is the perceptual attribute, and this perspective needs to be prioritized when design creativity is carried out.

Figure 1.

The dynamic graphics design factor is based on perceptual fluency

In the interview questionnaire, a total of 50 copies were sent out and 50 copies were successfully recovered, of which 46 were valid questionnaires and 4 were invalid questionnaires. Considering the personality characteristics of the people who use this intelligent design APP, this questionnaire was mainly used as a small-scale quantitative analysis for the study in the form of a scale questionnaire, so as to give a more precise assessment of the respondents’ scores.

The questionnaire data are shown in Table 4. The user ratings were based on a Likert scale, with 1-5 representing five grades of very favorable, relatively favorable, fair, relatively poor, and poor, respectively. The ratings of the surveyed users on the dynamic graphic design of this social APP are 3.57 points for perceptual attributes, 3.82 points for aesthetic attributes, 4.11 points for cognitive attributes, and 3.67 points for perceptual experience attributes, with the scores at the medium level and above. The score for the dynamic graphic design of the social app was calculated to be 75.68.

The realization of graphic dynamism is remarkable, and the application of dynamic graphics is far more valuable to design than traditional graphic design in terms of visual sensation and information communication.

Combining interviews and observations of the interviewees, it was found that the use of motion graphics has been accepted by the new media community. New media people hope that motion graphic design is not limited by media, and that it becomes a free, open, cross-regional, and multidisciplinary application design. It can be made more flexible by creating its own design according to the users’ needs. The advancement of intelligent technology and interactive technology also makes the new media people expect motion graphics to bring more different feelings to satisfy their needs of pursuing new things. Therefore, future design should pay more attention to the specific needs of each person and propose different solutions for various special situations.

Innovative graphic design has the following two main points: 1)

Focus on visual impact effect

Graphic design itself is a kind of art of visual communication, both for consumers and the design itself need to have visual focus, and the creation of visual focus is very important in visual expression. When people watch graphic design works, they first pay attention to the visual focus, and then pay attention to the shape and direction of the image, as well as the visual process of the shape changing with the change of color and movement trend.

And along with the development of the new media environment, dynamic graphic design has more advantages of visual communication than traditional graphic design, and the form of graphic image in the new media field is richer, combining more sensual experience based on humanized service, which makes the audience more comfortable.

Take the smart social app as an example, its own definition is to expand the new media crowd dating new media medium. Dynamic graphic design in the app emphasizes simplicity and clarity, the cell phone plane space is small, the dynamic performance and effect in the production of the logo need to maximize the concentration and simplicity. Overly complex and scattered dynamic effects not only cannot capture the audience’s visual focus, but also distract the audience’s perspective, leading to conveying incorrect or unimportant information. Therefore, motion graphic design needs to follow the principle of high expressiveness, i.e., not only to create visual focus but also to identify the whole. Under the premise of overall static, only the directional details of local action effects can grasp the visual focus from point to point.