Image retrieval of wool fabric. Part III: based on aggregated convolutional descriptors and approximate nearest neighbors search

Abstract

For sample reproduction, texture and color are both significant when the consumer has no specific or individual demands or cannot describe the requirements clearly. In this paper, an effective method based on aggregated convolutional descriptors and approximate nearest neighbors search was proposed to combine the texture and color feature for wool fabric retrieval. Aggregated convolutional descriptors from different layers were combined to characterize the wool fabric image. The approximate nearest neighbors search method Annoy was adopted for similarity measurement to balance the trade-off between the search performance and the elapsed time. A wool fabric image database containing 82,073 images was built to demonstrate the efficacy of the proposed method. Different feature extraction and similarity measurement methods were compared with the proposed method. Experimental results indicate that the proposed method can combine the texture and color feature, being effective and superior for image retrieval of wool fabric. The proposed scheme can provide references for the worker in the factory, saving a great deal of labor and material resources.

Research progress of content-based fabric image retrieval

“…These general features can usually be extracted from the first few layers of the CNN model. Zhang et al 61 used the VGG-16 model pre-trained on ImageNet as the feature extractor to characterize different kinds of fabric images. Salience maps of fabric images were extracted by aggregating the feature maps of convolution and pooling layers to obtain deep aggregation features.…”

“…This method can not only achieve identical performance but also avoids the inherent problems of model training. 61 For intuitive comparison, Table 2 summarizes different deep feature extraction methods in fabric image retrieval. Comparatively speaking, training CNN models from scratch requires a large number of labeled samples and sufficient computing resources, and the resulting model has strong generalization performance.…”

“…The feature dimension was reduced to improve the efficiency of fabric image retrieval. Subsequently, Zhang et al 61 used the ANNOY algorithm to process the deep aggregated features extracted by the CNN model, which can improve the retrieval efficiency and reduce the loss of retrieval precision. Figure 10 shows the process of constructing a feature index based on the ANNOY algorithm adopted in literature.…”

“…Figure 8. Activation regions and saliency maps of sampled feature maps extracted by the pre-trained model in literature 61. …”

“…Figure 10. Construction process of a feature index of Approximate Nearest Neighbors Oh Yeah (ANNOY) algorithm used in the literature 61. …”