JP2019028650A5 - - Google Patents

Claims (20)

An acquisition means for acquiring an input image whose element is a sensor value which is an output from the image element ,
Using identification unit has an identification means for identifying the input image,
The classifier, image identification apparatus characterized by learned based on the learning image to the sensor value output from the image element as an element. The discriminator includes a conversion unit that performs a predetermined conversion on an input image,
The image identifying apparatus according to claim 1, wherein at least the conversion unit of the identifier is learned based on the learning image. The conversion unit is
A first output of an output signal obtained by converting an input signal, which is a linear sum weighted by a weighting coefficient, of an absolute amount of brightness in each pixel of the input image and a scene feature extracted from the entire input image by a non-linear function Layers and
A second layer for outputting an output signal obtained by converting an input signal composed of a weighted linear sum of weighted coefficients of the scene feature and the output signal of the first layer by a linear function;
The image identifying apparatus according to claim 2, further comprising: The image identifying apparatus according to claim 3, wherein the scene feature is an image feature acquired from the entire input image. The image identification apparatus according to claim 3, wherein the scene features are image features acquired from the entire input image and imaging information when the input image is captured. The image classifier according to claim 1, wherein the classifier is trained based on the learning image and correct data indicating a classification result of the learning image. 7. The image identification apparatus according to claim 1, wherein the input image and the learning image each have a sensor value output from a CCD or a CMOS as an element. The discriminator is a discriminator obtained by adding the conversion unit to a discriminator learned based on first learning data including a developed learning image and correct answer data given to the learning image, any one of claims 1, characterized in that it is produced by learning based on the second learning data including the correct answer data by the sensor value of the image pickup apparatus is applied to the learning image and the learned image 7 1 The image identification device according to the item . The classifier, image identification apparatus according to any one of claims 1 7, characterized in that the neural network. The classifier according to claim wherein the feature extracting section for extracting image features by convolution of the multi-layer, and a classification unit that performs pattern classification from the extracted image feature, it is a neural network convolution including 9 The image identification device described in 1. The classifier, image according to claim 10, characterized in that the input to each layer of the feature extraction unit, at least one of the luminance value or distance information of each pixel of the input image as a weighted signal Identification device. Further comprising adjusting means for adjusting the conversion unit based on the identification result by the identifying means,
The image identifying apparatus according to claim 2 , wherein the identifying unit re-identifies the input image by using a classifier having the adjusted conversion unit. The input image is an image identification apparatus according to any one of claims 1 to 12, characterized in that it is a RAW image. 14. The identification means executes an identification task of classifying the input image into a plurality of classes, or an identification task of dividing the input image into a plurality of areas for each class. The image identification device according to item 1. A first learning means for learning the first discriminator based on the first learning image is the result of development of the input image to the sensor value output from the image element as an element,
Generating means for generating a second discriminator by adding a conversion unit that performs a predetermined conversion to an input image to the learned first discriminator;
Second learning means for learning the second discriminator on the basis of a second learning image which is an image having the sensor value output from the image element as an element ;
A learning device comprising: Acquire an input image with the sensor value that is the output from the image element as an element,
An image identification method characterized in that the input image is identified using an identifier learned based on a learning image having sensor values, which are outputs from image elements, as elements. A learned model that has been learned using a learning image having sensor values, which are outputs from image elements, as elements,
A learned model that outputs in order to identify the class of the input image according to the input of the input image that has the sensor value that is the output from the image element as an element. Learning a first discriminator based on a first learning image which is a result of developing an input image having a sensor value which is an output from the image element as an element,
A second discriminator is generated by adding a conversion unit that performs a predetermined conversion to an input image to the learned first discriminator.
A learning method characterized by learning the second discriminator based on a second learning image, which is an image whose elements are sensor values output from the image element. Program for functioning as an image identification apparatus according to any one of the computer claims 1 1 4. Program for functioning as a learning device according to the computer to claim 1 5.