A face sample cleaning method and system based on deep learning features

Abstract

The present invention is applicable to the fields of data mining and sample cleaning, and provides a face sample cleaning method based on deep learning features. The face sample cleaning method includes the following steps: A. Using deep learning technology to extract face image features ; B. Divide the samples that need to be processed according to the given number of processes and assign them to each process; C. Perform clustering processing on the corresponding face features in each process; D. In each clustering result Select the class with the largest number of images as the main class; E. Copy the images corresponding to the main class features to the target directory and retain the original file structure. The face features extracted by deep learning have low dimensions and very good discrimination ability. On the one hand, it can improve the cleaning accuracy of samples, and on the other hand, it can also save cleaning time; multi-process parallel design can make full use of the computing power of hardware, thereby improving The efficiency of the entire operation is reduced and time-consuming is reduced.

Description

technical field [0001] The invention belongs to the field of data mining and sample cleaning. The technology used is based on image processing and machine learning, and in particular relates to a face sample cleaning method and system based on deep learning features. Background technique [0002] Extracting facial features with deep learning: Deep learning is a branch of machine learning based on a family of algorithms that attempt to perform high-level abstractions on data using multiple processing layers that contain complex structures or consist of multiple non-linear transformations. An observation (such as an image) can be represented in a variety of ways, such as a vector of intensity values ​​for each pixel, or more abstractly as a series of edges, regions of a specific shape, etc. Instead, it is easier to learn tasks from examples (e.g., face recognition or facial expression recognition) using certain representations. [0003] DBSCAN (Density-Based Spatial Clusterin...

AI Technical Summary

Problems solved by technology

[0011] Since the feature extraction method is disturbed by noise, etc., and the effect of the classification model is often closely related to the accuracy of manual tuning parameters, the final accuracy is low, and the effect of distinguishing different faces is not good.

2. Due to the high feature dimension and the use of serial operation design, the whole process is inefficient and time-consuming. As the number of images increases, the time-consuming tends to increase exponentially.

3. When building the initial model, it is often necessary to manually mark part of the data

[0015] Based on the above-mentioned shortcomings in the prior art and other reasons such as single-threading, the implementation methods of the prior art are generally less accurate and time-consuming

Images