Intelligent garment matching recommendation method

Abstract

The invention discloses an intelligent garment matching recommendation method. The method comprises the following steps: establishing an upper and lower garment matching scoring model, marking the acquired garment images, respectively encoding each level of each element to be marked, marking as 0-35 feature vectors, constructing a convolutional neural network, then identifying and training the image data in the training set, stopping and fitting the function and storing the model when the loss function approaches convergence, performing multiple test experiments by using the data in the test set, and listing a confusion matrix according to a result to obtain an evaluation result. According to the invention, massive sample images and corresponding features and score data are learned througha machine, and finally, a system can recognize the dominant hue, style, profile and style characteristics of a single garment image which is input at will, and can automatically give a matching scorefor paired images which are input at will. For any input single garment image, the first ten garment images can be recommended according to the matching scores, a matching recommendation list is generated, and the first garment image is the optimal matching.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence applications, in particular to a method for recommending intelligent collocation of clothing. Background technique [0002] In February 2017, the "2017 Internet Fashion Consumption Trend Report" jointly released by Taobao's fashion interactive platform i Fashion and China Business News Center showed that Internet technology and fashion are becoming more and more closely related. The current large-scale e-commerce platforms such as Taobao and JD.com all provide search functions. On the one hand, the current search function is usually only for a single piece of clothing, and users need to search multiple times when they want to buy their favorite clothing collocation; The process of purchasing clothing is often time-consuming and labor-intensive, which greatly affects the consumer experience. On the other hand, there are a large number of users who do not have the time and energy t...

AI Technical Summary

Problems solved by technology

[0006] 1. At present, only the high-level semantic features (clothing style) of clothing images are often considered, and the importance of low-level visual features (such as color, silhouette, fabric texture, etc.) on clothing collocation is ignored;

[0007] 2. In the current research, only a simple set of positive and negative examples is used to classify pairs of clothing images, and then a large number of matching pairs and non-matching pairs are learned through machine learning. The matching is used as a binary classification problem to learn its potential matching rules.

Lack of collocation theoretical support, and it is inconvenient for users to compare the collocation of the same item with different collocations;

[0008] 3. Most of the current domestic clothing recommendation methods rely on the user's historical purchase / browsing records, which can easily lead to the recommendation results being too similar to the user's past purchase styles, and it is impossible to tap the styles that users may potentially like

Images