关键词: 纹理特征, 数字纹理色卡, 图像处理, 机织物, 色差, 纹理相似度

Abstract:

Objective In the textile and apparel industry, color cards are an important communication tool between design and production. However, traditional color cards fail to represent color of textured fabrics accurately, leading to visual discrepancies between standard colors and actual fabric appearances. In order to tackle the visual difference between a single color in the traditional color card and the real fabric color with texture features, this study aims to develop a digital texture color card by fusing texture templates with specified color values through a novel algorithm.

Method In the Lab color space, an algorithm of digital texture color card was proposed to adjust fabric brightness by subtracting the average gray value of L channel and superimposing the target color value of color channel. An Epson Perfection V19 scanner was used to capture fabric images at 1 200 dpi, and the images were then de-noised using median filter and geometrically corrected by fast Fourier transform and Hough transform to obtain the texture template. According to the texture color card algorithm, fabric digital texture color cards were made and human visual perception experiments designed. The influence of fabric color on texture during scanning was analyzed, and the digital texture color card was evaluated from two aspects, i.e. color difference and texture similarity.

Results Four fabric images of different colors and the same texture structure were collected by using the scanner, numbered 1 to 4 according to the brightness of the fabric from low to high, and the related image processing and texture color cards were completed in MatLab R2023b. As the brightness of the fabric increases, the distribution range of gray values decreases. It was found that the gray histogram of fabrics 1 to 3 was high in the middle and low on both sides. For fabric 4, due to its high brightness, its gray histogram showed a left-low and right-high trend indicating its detail texture is not clearly visible in the scanned images. By calculating the gray co-occurrence matrix and related mean values in the four directions of the fabric, it was concluded that with the increase of fabric brightness, the energy and correlation increase, and the texture rules were uniform and the directionality was obvious. Under the above conditions, the entropy and contrast were reduced, the scanning texture was more regular, the image appeared smoother, and the texture change was not significant. CIEDE2000 color difference formula was used to calculate the color difference between the texture color card and the original fabric image to evaluate the color proximity. The results showed that the mean value of CIEDE2000ΔE between the texture color card and the original fabric was between 0.05 and 0.69, and the color difference was small. In order to verify the rationality of the method of making the texture color card, histogram cosine similarity (HCS) and structural similarity (SSIM) were used to verify the similarity between the texture color card and the scanned fabric image. The results exhibited that the mean value of texture color card HCS was above 0.90, and the similarity of color distribution was high. From the perspective of structure, the mean value of SSIM of fabrisc 1, 2 and 3 exceeded 0.8 and the texture similarity was high, while the mean value of SSIM of fabric 4 was low and the standard deviation was high relatively. This method is suitable for making fabric texture color cards with moderate brightness. In the experiment of human visual evaluation, the means of subjective scores were distributed in the range of 0.23 to 0.83, corresponding to the perceived level of visual difference ranging from "no difference" to "barely perceptible difference". Futhermore, the results of subjective evaluation were highly consistent with those of objective evaluation.

Conclusion According to the texture color card algorithm, a fabric digital color card with texture features is made to realize the presentation of different colors on the texture. It is found by scanning fabric samples with different texture images and different brightness. The higher the brightness, the smoother the scanned texture. The color difference of the texture color card is small, and the color value is consistent with the scanned image. The mean values of HCS are close to 1, and the color distribution is similar. For texture color cards with medium brightness values, SSIM mean values are above 0.8, and texture similarity is high. The results achieved high fidelity (HCS>0.90, SSIM>0.80) for test samples, demonstrating robustness for medium-brightness woven fabrics. Fabrics with extreme brightness deviations (overly high or low) should be paired with texture templates of comparable brightness levels. In the subjective evaluation, texture affects the visual perception of color. The cosine similarity index based on HCS histogram is closer to the visual characteristics of human eyes in the evaluation of texture perception. This method is currently suitable for medium-brightness fabrics. In the future, we will focus on optimizing the algorithm to improve the color visual effect with a wider range which is expected to be suitable for fabrics with different color and texture, and create a digital database of fabric texture color cards.

Key words: texture feature, digital texture color card, image processing, woven fabric, color difference, texture similarity