Color feature extraction method and clothing retrieval system based on classified clothing

Abstract

The invention discloses a color feature extraction method and clothing retrieval system based on classified clothing. The color feature extraction method comprises the following steps: obtaining a clothing image; transforming the color of the clothing image into a HSV (Hue, Saturation, Value) space; removing the background of the clothing image; calculating the H component color histogram of the clothing image, and carrying out N-order quantification on the H component color histogram; searching a component with a largest proportion in the H component color histogram as a maximum H component peak value; adopting a peak value judgment method based on a threshold value to search H component peak values which meet a condition; and according to the sum of the H component peak values, carrying out clothing image classification, and independently correspondingly selecting a corresponding clustering number and an initial clustering center by aiming at each category of clothing images. By use of the method, according to the classified clothing, the clustering number and the initial clustering center are determined, a more stable main color feature value can be extracted, and a calculated amount and calculation time during color feature extraction can be effectively reduced. When the method is applied to clothing image retrieval, the precision ratio and the recall ratio of retrieval can be effectively improved, and a retrieval result is more stable.

Description

technical field [0001] The invention relates to the technical field of clothing image retrieval, in particular to a color feature extraction method based on classified clothing and a clothing retrieval system. Background technique [0002] Traditional clothing image retrieval is based on text, and users search by inputting keywords describing clothing. This method is called text-based image retrieval (TBIR). However, with the rapid development of the clothing online shopping market, the number of online clothing has increased rapidly. The traditional TBIR system has been unable to meet the needs of users due to its strong subjectivity, limited text description information, and low search efficiency. Therefore, content-based image Retrieval (CBIR) technology came into being. At present, the underlying visual features (color, shape, texture, etc.) of the image are mainly used to describe the content information of the image. [0003] Color is the most stable visual feature o...

AI Technical Summary

Problems solved by technology

The existing method for extracting the main color of clothing images is mainly the k-means clustering algorithm, but there are two shortcomings when the traditional k-means clustering algorithm is applied to clothing retrieval: first, in the traditional algorithm, the number of clusters of images is artificial a fixed value

In fact, the number of main colors to be extracted for different types of clothing is often different, so it is not appropriate to use a fixed value

Second, traditional algorithms usually randomly designate initial cluster centers, which will lead to unstable retrieval results and affect retrieval precision and recall

Moreover, since the randomly designated initial cluster centers do not consider the color characteristics of the image itself, the algorithm needs to perform a large number of iterative operations to find the real cluster centers, which requires a large amount of calculation and time-consuming calculation.

Images