Improving color reliability of digital textile images via optimized acquisition and preprocessing

Abstract

As textile industries move toward greater digitalization and automation, accurate and reproducible image-based analysis of textile surfaces has become increasingly important across research, manufacturing, and digital commerce. In this study, we propose an integrated imaging and preprocessing framework that significantly improves the color reliability of digital textile images, thereby enhancing the precision of textile image analysis across a wide range of applications. We compared the extent of color distortion in digital images of samples from 11 different textile under different background and surrounding colors (white or black). We evaluated the effectiveness of image fusion as a preprocessing step by comparing the color reliability and image quality of non-fused images. We then derived optimal settings for image acquisition environments and preprocessing strategies for digital images of textile surfaces. To validate the proposed guidelines, we compared the color distortion in digital images corrected using only conventional color correction, and to digital images obtained using our proposed guidelines before color correction. Our findings demonstrated that applying appropriate physical environments and image preprocessing significantly increases the color stability and reliability of digitally represented textile colors, yielding an average improvement of 21.2% in their visual fidelity to the original fabric.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-30857-x.

Introduction

Digital image analysis (IA) has become an indispensable tool in textile research and industry. However, surfaces structures in textiles are diverse and irregular, so image analysis (IA) of textiles is a challenge. Textiles vary in surface properties such as gloss, transparency, and texture, and these variation can influence the color of the textiles in digital images¹. To address this problem, artificial intelligence (AI) has been integrated with textile IA for various purposes^2–6. For instance, computer vision studies on content-based textile image retrieval^4,7 and product classification^8,9 have provided effective solutions to enable automation and digitalization of online textile commerce. Digital IA has also been used to detect fabric defects^10,11 and to evaluate dyeing properties^12–14; these advances have increased the efficiency and sustainability of textile manufacturing processes. These results demonstrate that AI-integrated textile digital IA has potential applications in various domains^2,15.

Most IA techniques rely primarily on colors as key features of digital images^16,17. Color reliability—the ability to consistently represent real-world textile colors in digital images—is therefore critical for ensuring the reproducibility and accuracy of textile IA results. For example, digital IA has been used to calculate detergency by deriving the ratio of absorbance K to scattering S from pixel color data before and after textile washing, thereby improving the precision of stain removal assessment compared to traditional methods¹⁷. Similarly, a study on the dyeability of polyester knitted fabrics analyzed dye diffusion and color depth by examining the surface color of dyed samples captured in digital images¹⁸. To increase color reliability and experimental reproducibility, prior studies^17–19 have suggested preprocessing methods such as image fusion or digital color correction using standard color checkers. However, relatively little attention has been given to optimizing the physical environment during image acquisition, despite its potential impact on color distortion in textile images.

Psychological studies have shown that human color perception is affected by background lightness and contrast with the target object^20,21. Similarly, digital cameras are influenced by these factors due to white balance settings and autofocus systems that are sensitive to contrast^22,23. In particular, when the color contrast between the sample and background is low, focus accuracy may decrease, resulting in increased image noise. Moreover, researchers working on textile property evaluation using textile IA have frequently observed, through visual inspection, that background colors often introduce significant color distortion in the captured textile images, leading to inaccurate color representation. Therefore, to minimize color distortion and increase the quality of digital images, the influence of physical factors such as background colors, must be studied experimentally.

Additionally, the three-dimensional surface geometry of textiles complicates consistent color rendering in digital images. Image fusion techniques—particularly those using stereo-photometric methods—can enhance color reliability by integrating multiple images captured under varied lighting angles^1,16,17,19. However, best practices for implementing these methods remain fragmented.

Thus, this study aimed to identify optimal image acquisition environments and preprocessing strategies for textile IA. As a research group developing a digital image–based textile detergency evaluation method, we used artificially soiled samples, whose colors change after washing. To ensure consistent imaging regardless of such changes, we sought to establish standardized acquisition conditions and preprocessing steps. As illustrated in Fig. 1, we first examined the effects of background and surrounding colors on camera autofocus performance and color distortion, to determine the most suitable background and inner wall colors for textile imaging devices. We also validated the effectiveness of image fusion as a preprocessing step and investigated optimal lighting conditions for image fusion. Finally, we compare the quality of digital images obtained under conventional color correction methods versus those captured and processed according to our proposed guidelines, and thereby verified the effectiveness of the suggested approach.

Fig. 1.

Schematic illustration of the study investigating the optimal digital image acquisition settings and preprocessing process for analysis of digital images of textiles.

Experimental

Materials and pretreatment

This study was conducted using white polyester fabric (Young Poong Filltex, Co. Ltd., South Korea) and five artificially-soiled fabrics of different colors (Testfabrics Korea, Inc., South Korea) as samples (Table 1). The goal was to identify image-acquisition settings that remain effective even after washing, which alters the colors of artificially-soiled fabrics. The artificially-soiled fabrics were washed in 500 mL of a 0.2% detergent solution (Persil, Henkel AG & Co., Germany) at 60 °C for 30 min with continuous stirring: three samples of each artificially-soiled fabric were placed in a beaker with detergent solution and stirred at 150 rpm using a magnetic stirrer.

Table 1.

Sample codes, characteristics and color names using practical color coordinate system (PCCS).

Image acquisition

A stereo-photometric device (Figure S1, Apparelbase Co. Ltd., South Korea) was used for image acquisition. This device consists of a black sample stage and white acrylic inner walls, and it allows independent control of LED D65 standard (6500 K) illumination from eight different angles (Figure S1). Each LED source was powered by a constant-current power supply to ensure stable output. The D65 LED source produced an illuminance of 3,352 ± 10 lx on the sample plane when only one directional LED panel was activated in the photometric system (measured three times using a LUAZ-280 digital lux meter). This illuminance level falls within the typical range used in textile color evaluation environments (approximately 1,000 ~ 5,000 lx).

Uncontrolled variations in illumination intensity can affect the RGB values and Lab conversion, thereby influence -based color reliability²⁴. To verify the uniformity of illumination within the imaging box, a standard gray color checker was used to evaluate the illuminance distribution across the sample plane²⁵. Measurements confirmed that the lighting within the lightbox remained spatially consistent.

To modify the colors of the sample stage and inner walls, black or white paper was cut to the appropriate size and attached. For capture of digital images, a 5 cm × 5 cm fabric sample was placed at the center of the sample stage, and a camera was mounted above for image acquisition. Images were captured under two lighting conditions: (1) with only one light source turned on Figure S1, L1–L8), and (2) with all light sources turned on simultaneously (Figure S1). The distance and illumination angle between the light sources and the textile surface were fixed throughout all image acquisitions to maintain consistent geometric lighting conditions.

During image acquisition, the camera-to-sample distance was fixed at 25 cm. However, slight variations in sample surface height – caused by weave structure – occasionally shifted the focal plane. To ensure optimal sharpness, the autofocus (AF) function of the CCD camera (Canon EOS 450D) was activated, and the flash was turned off. The camera sensor has a size of 12.2 million pixels, with a pixel size of 5.19 μm, which is sufficient to capture the textile structure and irregular stains on the fabric surface. The lens used was an EF 24–70 mm zoom lens, and all images were taken in autofocus mode with an object distance of 25 cm.

Image preprocessing

Fusion

To evaluate the effect of image fusion, images captured under different lighting conditions from various angles were combined into a single image by using the discrete wavelet transform (DWT) image-fusion method^17,19. To identify effective lighting conditions for image fusion, two sets of image fusion were performed: one using four-directional lighting (0°, 90°, 180°, 270°) and another using eight-directional lighting (0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°). The region of interest (512 × 512 px) was cropped from the captured original image, the a single-level DWT with a symmetric image boundary handling method and the ‘Haar’ wavelet form was applied. After DWT had been performed on the four (or eight) cropped images, the wavelet coefficients were averaged and combined to generate the final fused image by using inverse discrete wavelet transforms (IDWTs) (Figure S2). For the image fusion process, the ‘dwt2’ function in MATLAB software (R2024a) and was used.

Colorimetric calibration

For color correction of digital images, the X-Rite Macbeth 24 standard color chart was placed within the frame along with the fabric sample during image acquisition. The L*a*b* values from the CIELAB color system of the 24 color patches were extracted from the digital images, and a correction-matrix method was applied to minimize the difference between the extracted L*a*b* values and the original patch values. This method adjusts the image color values to closely match the standard reference colors, and thereby increases the color accuracy of the entire image. Specifically, the transformation between the measured L*a*b* values and the reference L*a*b*values was modeled as a linear color transformation matrix M:

1

where represents the known reference L*a*b* values from the color chart, represents the measured L*a*b* values extracted from the digital image, and is the color-correction matrix, which is computed using least squares regression to minimize the color difference

2

Color verification

To verify the effectiveness of the proposed image acquisition environment and preprocessing in reducing color distortion in textile digital images, color difference evaluation methods were used.

Color difference with reference L*a*b* of color chart

The X-Rite Macbeth 24 standard color chart was included in fabric image acquisition, and under different imaging environments and preprocessing conditions, the original L*a*b* values of the color patches were compared with the L*a*b* values extracted from the digital images. The color difference , was calculated using the CIE76 color-difference formula (Eq. 3). A low was considered to indicate low color distortion in the digital image.

Color difference with measured L*a*b* by colorimeter

The L*a*b* values of each fabric sample were measured using a colorimeter (Minolta CM-2600d) and compared with the L*a*b* values extracted from digital images captured in different environments. To prevent the influence of fabric thickness on the measurement, four identical specimens were overlaid when the colorimeter was used²⁶. The surface color was measured at four random locations on each specimen, then averaged. The color difference was calculated as

3

where represent the and values of the reference color or color measured using the colorimeter, and represent the and values extracted from digital images.

Results and discussion

Color of fabric samples before and after washing

The color of white polyester fabric and artificially soiled fabrics before and after washing (Table 2) was observed using digital images and a colorimeter. Washing of all artificially-soiled fabrics increased their luminance and decreased their saturation. Visual inspection of the samples identified that the texture resulting from the three-dimensional surface characteristics of interwoven yarns influences the perceived color, and this characteristic was also reflected in high-resolution digital images. In contrast, color measurements obtained using the colorimeter excluded the texture effects such as shading caused by the interwoven yarns, and therefore provided color information that significantly deviated from the actual perceived color.

Table 2.

Color changes of samples before and after washing.

Additionally, the degree of uniformity in washing varied, depending on the type of contaminant. In particular, for fabric that had been artificially soiled using carbon black (COO), and therefore contained solid particles, the washing solution did not completely disperse the contaminants, so in some particles remained trapped within the fabric structure. These differences in washing results depending on the type of contaminant were clearly observable both visually and in digital images. However, the colorimeter, with its measurement aperture diameter of 3 mm, had a significantly lower resolution than digital image and was unable to capture the fine variations or non-uniformities in fabric-surface colors.

Effects of background and surrounding colors on color distortion in digital images and camera autofocus function

were calculated by comparing the original digital image colors of fabric samples to the colorimeter measurements (Eq. 3) with the background being either white or black. Unprocessed digital images without color correction showed significant compared to colorimeter values in both background conditions (Fig. 2). However, for most samples, was lower with a white background than with a black one. This indicates that color distortion was less pronounced when the background was white than when it was black.

Fig. 2.

Color differences between colors measured using a colorimeter and those extracted from digital images of samples under different background and interior wall colors during digital image acquisition.

During digital image capture, excessive brightness or increased color saturation on a black background may lead to color distortion due to factors such as camera-sensor characteristics, image-processing algorithms used, and optical-contrast effects^27–29. First, most digital cameras apply gamma correction, which can excessively increase contrast between light and dark areas, particularly against a dark background^28,30. Additionally, cameras use white-balance algorithms to adjust color temperature according to the shooting environment. When the background is dark, the camera may emphasize chromatic colors to prevent the image from appearing too dim, but this process can lead to excessive distortion of certain colors³¹.

When both the background and interior walls were white, color distortion was reduced, but the digital camera’s autofocus function failed to operate for the UT and CRW (W) samples, so image capture was impossible (Fig. 2). This problem may have occurred because of the low contrast between the white background and the UT and CRW (W) samples, which have high luminance, low saturation. The camera’s autofocus algorithm could not operate correctly in this condition, hindered the proper functioning of autofocus. Compared to other washed artificially soiled fabric samples, the CRW (W) had the lowest contrast (20.68) with the white background (Fig. 3a). Additionally, when the interior walls of the imaging box were white, their high reflectance increased the amount of light scattered from multiple angles (Fig. 3b, left). As a result, this scattered light entered the camera lens, and the resulting lens refraction interfered with calculation of the focal length; this problem may have further contributed to the failure of digital camera’s autofocus functionality.

Fig. 3.

(a) Contrast between the white background and fabric samples; (b) effects of the sample stage and interior wall colors on light reflection and lens focus.

This can be described as follows (Fig. 4): for sample groups that had a Euclidean distance (ED) ≤ 10 or less from white () on the luminance-saturation graph, the autofocus function may not work properly when both the background and interior walls are white. To address this problem, the interior walls of the imaging box were replaced with black while the background was kept white (Fig. 3b, right). This adjustment enabled the camera’s autofocus function to operate normally, and allowed successful acquisition of digital images for the UT and CRW (W) samples.

Fig. 4.

Luminance and saturation of white and black paper and fabric samples.

In this study, black and white backgrounds were employed to represent two extreme conditions of reflectance contrast. This approach was intentionally designed to examine the maximum potential range of background-dependent color distortion that could occur during digital image acquisition. The results obtained under these contrasting conditions provided valuable insight into the extent to which background reflectance may influence image exposure and measured color values.

Nevertheless, a neutral gray background could serve as a balanced and standardized alternative between these extremes. Previous studies and imaging standards have suggested that a neutral-gray background can help minimize background-induced bias and stabilize exposure levels by providing an optically neutral environment under diffuse illumination conditions³². Other research has also indicated that neutral-gray surroundings may reduce scattering and halo effects that interfere with accurate color measurement and -based color difference computations³³. As such findings imply potential advantages of using a neutral-gray background, further validation is needed to determine its influence within the context of textile imaging systems. Future studies will therefore examine this factor in greater detail to optimizing imaging conditions for digital image-based textile color evaluation.

Effect of image fusion using a stereo photometric system

Use of fused images from a stereo photometric system can effectively reduce color distortion in textile digital images because textiles have a three-dimensional surface structure formed by interwoven warp and weft yarns^17,19. In this study, a stereo photometric device with an eight-angle light source (Figure S1) was used to compare the digital color of non-fused images and those fused from four-angle and eight-angle light sources (Fig. 5a). The digital K/S values differed clearly between fused and non-fused images (Fig. 5b). Additionally, the color variations, as indicated by the error bars in the graph, were smaller in fused images than in unfused images. Fusing images from separately captured textile images under photometric lighting minimizes color distortion on textile surfaces, and increases the accuracy and stability of color values¹⁶. This improvement occurs because combining multiple images illuminated from different directions helps mitigate shadows and highlight effects that may occur with a single light source, and there by increases the uniformity of illumination across the surface and yields a consistent color representation.

Fig. 5.

(a) Digital images of non-fused image and fused images (above) and illustration of image acquisition process with single and multi-direction light sources (below); (b) Digital K/S values (K/S_D) of non-fused image and fused images with 4-direction (Fused (4)) and 8-direction (Fused (8)) light sources.

Textile surfaces reflect light anisotropically because of the three-dimensional surface structure that results from the intersection of yarns, and from the fine undulations caused by fiber strands¹⁹. Consequently, use of only one light source can exaggerate or diminish certain reflections, and thereby cause color distortion. In contrast, in fused images from multiple cameras, reflections from the entire surface are integrated; the process reduces anisotropy and helps to increase the reliability of the reproduction of the true color of the textile³⁴. However, increasing the number of light angles from four to eight for image fusion did not significantly affect on image color (Fig. 5). Therefore, this study indicates that a four-angle light source in the stereo photometric system is sufficient to effectively and efficiently achieve color stability by image-fusion preprocessing.

Effect of proposed image acquisition environment and preprocessing on reducing color distortion

The experimental results indicate that the optimal image acquisition environment and preprocessing method to increase image quality while ensuring camera autofocus functionality and minimizing color distortion is (Fig. 6):

Fig. 6.

Proposed image acquisition setup and image preprocessing for textile digital IA.

• I.A four-directional stereo photometric imaging environment with a white background and black interior walls.
• II.Fuse four individual images captured under a four-directional light source.
• III.Calibrate color by using a correction matrix with a standard color chart.

of the standard color chart appeared closer to its original color when the proposed method was used than only conventional color calibration was used (Fig. 7). The effectiveness of this approach was validated using a standard color chart rather than textile samples because the original L*a*b* values of the color chart could be objectively compared with the L*a*b* values from digital images to calculate and assess color distortion more accurately than by using textile samples.

Fig. 7.

Color differences between L*a*b* values from digital standard color chart images and reference L*a*b* values based on image acquisition environment and image preprocessing.

Among the 24 colors in the standard color chart, the 10 most similar to the colors of the washed and unwashed artificially soiled fabric samples were selected for color verification (Fig. 7). Control images were captured using a conventional imaging box with a black background and black interior walls, followed by color calibration. In contrast, test images were captured and preprocessed according to the proposed guidelines. While values varied across all images, they were generally smaller in the test images than in the control images. In a few cases—such as 40% Gray, 80% Gray, and Pink—the was slightly higher in the test images, but the differences were negligible. Conversely, images captured using the conventional method (control images) occasionally exhibited substantial color distortion, particularly in the cases of “Card White” and “Primary Red.”

These results suggest that the proposed guidelines provide a more robust and universally applicable solution for capturing digital images of textiles with diverse colors. The improvement in color reliability and image quality was also supported by visual assessments conducted by the researchers. Subjective comparisons between the actual sample colors and their digital representations aligned with the -based quantitative evaluations, indicating that the proposed approach significantly enhances the color fidelity of textile IA. To support broader application in high-precision textile color analysis, further validation through panel-based visual assessments or spectro-radiometric measurements is recommended.

These outcomes may be attributed to gamma correction algorithms applied by digital cameras, which tend to exaggerate contrast in dark background environments^28,30. Additionally, in low-light conditions, white-balance adjustments intensify chromatic colors to brighten the images, but this process can yield exaggerated and overly-vivid colors³¹. By using a white background, such excessive color adjustments by the digital camera are prevented, and maintaining black interior walls ensures that autofocus functionality is not lost. Furthermore, the three-dimensional surface structure of textiles can reduce the reliability of digital color measurements, so stereo photometric image fusion preprocess can increase the color consistency of digital images of textile. The color, pattern, and surface structure of textiles vary immensely, so IA that follows the proposed guideline (Fig. 6) can provide a reliable and stable method to obtain consistent digital color representations across any textile sample.

To provide context for the robustness and applicability of the proposed approach, we compared it with representative methodologies reported in recent literature (Table 3). Previous research, such as Rout et al. (2022³⁵, 2023³⁶ and Wang et al. (2020)³⁷, has made contributions to enhancing textile color accuracy through calibrated illumination and spectral reconstruction techniques. Building upon these foundations, our method offers a complementary direction – achieving reliable and spatially uniform color representation through optimized image acquisition and DWT-based fusion.

Table 3.

Comparison of representative methodologies for textile color analysis.

Study	Methodology	Key features	Limitations
Rout et al. (2022)35	Calibrated sphere imaging system (I)	Controlled multi-directional D65 illumination with integrated spectral–texture analysis	Requires precise geometric calibration and closed imaging sphere; limited adaptability to diverse sample sizes
Rout et al. (2023)36	Calibrated sphere imaging system (II)	Enhanced spectral correction and texture correlation analysis under uniform lighting	System complexity and maintenance cost can restrict broader applicability
Wang et al. (2020)37	Spectral reflectance reconstruction for color calibration	Accurate RGB-to-reflectance conversion using spectrophotometric reference data	Relies on spectral instrumentation and controlled laboratory conditions; not easily transferable to general imaging setups
Present study	Optimized acquisition + DWT-based image fusion	Stereo-photometric digital imaging; perceptually uniform CIE Lab color analysis; -based reliability evaluation	Current stereo-photometric configuration requires further adaptation for broader applicability

Conclusion

Our work presents a systematic investigation into the influence of physical imaging environments and preprocessing techniques (e.g., DWT-based image fusion) on digital color distortion. Based on extensive experiments using 11 different textile samples, we propose an image preprocessing guideline that achieves high fidelity between the actual fabric colors and their digital representations—beyond what conventional color calibration methods can provide. The work provided findings as below:

1. After washing, the five artificially soiled fabrics showed increased luminance and decreased saturation due to the removal of contaminants. Visual inspection of the digital images revealed that fabric surface texture influenced color appearance and that residual contaminants were unevenly distributed across the samples. These color characteristics were not accurately captured by colorimeter measurements. This highlights the limitations of widely used colorimeters in detecting spatially non-uniform stains or surface-dependent color artifacts—issues that can be effectively addressed through properly optimized digital imaging.

2. To obtain high-quality digital images with reduced color distortion and reliable focus performance, it was necessary to appropriately adjust the interior color of the lighting box. The background and surrounding colors inside the imaging box had a significant impact on internal light reflection and exposure balance, emphasizing the importance of controlling these factors to ensure stable illumination and accurate color rendition.

3. The novelty of this study lies in the application of discrete wavelet transform (DWT)-based image fusion within a stereo-photometric framework as a preprocessing step prior to color correction. Because textiles have a three-dimensional surface structure formed by interwoven yarns, fusing multiple images captured under multi-angle lighting produced more stable digital color representations than a single non-fused image provided. However, further segmenting the light source angles from four to eight directions did not yield significant benefits. A four-directional light source was sufficient to achieve color consistency of textile digital images.

By exploiting these findings, a three-step guideline was established: (1) Use an imaging environment that minimizes scattered light, which can interfere with accurate color measurement, (2) Capture individual images under a four-directional light source, and fuse them, and (3) Perform color calibration using a standard color chart. Digital images obtained using the proposed guideline yielded digitally represented colors that more closely resembled the actual fabric colors than those obtained through conventional color calibration alone, resulting in an average improvement of 21.2% in color fidelity.

This study confirms that appropriate image acquisition environments and preprocessing procedures enhance the reliability and effectiveness of color correction in digital images of textiles. Although the research was initially conducted to identify optimal settings for a digital image-based method for evaluating textile detergency, the findings have broader practical and research implications for a wide range of textile IA applications. However, this work has certain limitations: the stereo-photometric system used may have limited general applicability, and the potential influence of neutral-gray backgrounds was not experimentally validated. Therefore, future studies should aim to verify the effects of neutral backgrounds and extend the current framework to more universal and accessible imaging setups to enhance its practicality and reproducibility.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Yoonkyung Cho conceptualized and designed the study, performed the experiments, collected and analyzed the data, and drafted and reviewed the manuscript.

Funding

This work was supported by Incheon National University under Grant Number 2024 − 0285.

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.