First, gradient image is constructed by using direction templates convolve with image.

首先利用方向模板对图像进行卷积，求得梯度图像。

Two-value iris template will be achieved by compare every point of convolve image with zero.

对虹膜区域进行二值化编码，获得虹膜特征模板。

We then convolve x_image with the weight tensor, add the bias, apply the ReLU function, and finally max pool.

我们把x_image和权值向量进行卷积，加上偏置项，然后应用ReLU激活函数，最后进行max pooling。

They could be used directly as a projection basis to reduce dimensionality, but it needs to consider how to convolve them properly with the data you want to project.

他们可以直接使用作为预测的基础上，以减少维数，但它需要考虑如何卷积的数据要妥善项目。

The HVS function, HVS function is inversely Fourier transformed in frequency domain, and the method is realized by calculating some value of the function in spacial domain as convolve nucleus.

根据HVS函数的特点，对频域中HVS函数进行反傅立叶变换，将空间域中该函数的部分函数值作为特定的模板参数，并与原图像进行卷积来实现边缘检测。