Multiscale face detection method and computing equipment

Abstract

The invention discloses a multiscale face detection method and computing equipment for executing the method. The method comprises the following steps: a first-scale image is input into a preset convolutional network, and a first-scale face predication area and a second-scale to-be-detected area are generated after convolutional processing; an image of the second-scale to-be-detected area is cut from a second-scale image and input into the preset convolutional network, and a second-scale face predication area and a third-scale to-be-detected area are generated after convolutional processing; animage of the third-scale to-be-detected area is cut from a third-scale image and input into the preset convolutional network, and a third-scale face predication area is generated after convolutionalprocessing; a face detection result is obtained by combining the first-scale face predication area, the second-scale face predication area and the third-scale face predication area.

Description

Technical field [0001] The invention relates to the technical field of image processing, in particular to a multi-scale face detection method and computing equipment. Background technique [0002] Face detection means that for any given frame of image, a certain strategy is used to search it to determine whether it contains a human face. If it contains a human face, the position, size and posture of the human face are returned. Generally, a Rectangular frame to frame the face to represent the detected face. However, in practical applications, many complex conditions such as different scene changes, occlusions, illumination changes, and face scale changes are often faced, making the speed of face detection relatively slow. Especially when real-time detection needs to be implemented on a mobile terminal, face detection is very difficult. [0003] In order to detect faces at different scales, the image is usually scaled to multiple scales, and the face is detected separately at each...

AI Technical Summary

Problems solved by technology

The amount of calculation of this kind of multiple scales is very large, especially for the case where the face in the image is very small (referred to as "small face"), it is necessary to set the size of the input image to be larger, but this will increase the calculation consumption. Time

With the development of neural network technology, more and more algorithms use convolutional neural network for face detection. In this case, when the above face detection scheme is applied to convolutional neural network, with the input image Increase, the detection time will increase exponentially, and the calculation is quite time-consuming

At the same time, in many application scenarios, the image actually contains few small faces. In order to detect the small faces that may not exist, it is very undesirable to consume the calculation amount for the whole image.

Images