Research And Implementation Of Fabric Pattern Retrieval Algorithm

With the textile industry constantly evolving and people’s diversified pursuit of fabric patterns,there are more and more types of textile fabric patterns.For textile enterprises and fabric pattern designers,there is an urgent need for a method that can quickly retrieve the same or similar fabric patterns with the desired patterns in the massive pattern database to improve production efficiency.Currently,many textile companies use experienced manual retrieval or keyword-based retrieval,these two retrieval methods are highly subjective and often require a lot of time and effort,and it is difficult to ensure the accuracy of the retrieval.Therefore,we need a digital and intelligent way to achieve efficient retrieval of fabric patterns to meet the requirements of enterprises for fast and accurate retrieval of fabric patterns.The existing retrieval methods of fabric pattern are mainly based on single feature or fused feature,which can not meet the different needs of users in retrieval,such as the color similarity,texture similarity or color and texture similarity of the pattern.Based on the above problems,here are three practical requirements from the perspectives of textile fabric companies and pattern designers,this paper mainly does the following work:(1)Aiming at the demand of color similarity retrieval focusing on fabric patterns,a retrieval algorithm based on the combination of pattern main color and its aggregation region and block color moments is proposed.Firstly,the colors of the pattern were quantized into 9 intervals using a non-equidistant quantization method,next,an enhanced clustering algorithm is utilized to extract the primary colors of the pattern as global features;At the same time,cluster region features are extracted based on the primary colors of the pattern to describe the spatial distribution characteristics of the main colors;Then the color moments of the block pattern are extracted to represent the local features to improve the ability to distinguish local colors.Finally,we measure the similarity of the three features respectively,and then weighted the sum to obtain the overall similarity.The experimental results indicate that the method performs better than other color feature extraction methods of the same category,and the average precision in the fabric pattern database reaches 85.8%,which can effectively retrieve patterns with similar color styles.(2)Aiming at the requirement of texture similarity retrieval focusing on fabric patterns,a retrieval algorithm combining principal direction adaptive threshold LBP and Gabor wavelet features is proposed.From both global and local perspectives,the main direction adaptive threshold LBP algorithm is used in the spatial domain to extract local texture features.The texture features extracted by this algorithm have strong anti-noise ability,rotation invariance and other characteristics;Then,Gabor wavelet transform algorithm is used to extract multi-directional and multi-scale texture information as global features in frequency domain;Finally,a weighted fusion is performed on the two types of feature similarities.Experimental results demonstrate that the proposed method is capable of retrieving fabric pattern with emphasis on texture similarity,and the average precision in the fabric pattern database reaches 72.5%.Compared with other texture feature extraction algorithms,this algorithm performs better,demonstrating its superiority.(3)Aiming at the requirement of joint similarity retrieval focusing on fabric pattern color and texture,a hierarchical retrieval algorithm is proposed,which combines high-level semantic features and low-level visual features.Two hierarchical retrieval strategies are designed for general fabric patterns and special quasi-regular pattern patterns.First of all,the depth feature of VGG16 model is used to achieve preliminary retrieval,and then for general fabric patterns,perform fine-grained retrieval using the color and texture feature fusion algorithm proposed in this article,while for quasi-regular patterns,a global color feature and Fourier texture feature fusion algorithm is designed for fine retrieval.After experimental verification,the algorithm put forward in this article has higher average retrieval accuracy on both fabric pattern dataset and quasi-regular pattern dataset,with m AP of 90.28% and 89.96% respectively,which is superior to other algorithms.