Processing method for performing process on image and processing device using the processing method

1. A processing device comprising,

an input unit that receives an input of an image subject to a process;

a processing unit that subjects the Image Input to the input unit to a process of a convolutional neural network in which a fully connected layer is excluded; and

an output unit that outputs a result of the process in the processing unit,

wherein the convolutional neural network in the processing unit, includes a convolutional layer and a pooling layer;

wherein a filter of the convolutional layer in the convolutional neural network in the processing unit is trained to learn the result of the process having a 1×1 spatial dimension,

wherein a size of the image input to the input unit is larger than a size of an image for learning input to the convolutional neural network when the filter of the convolutional layer is trained for learning, the image for learning corresponding to the result of the process used to train the filter of the convolutional layer Air learning,

wherein the convolutional neural network in the processing unit performs a downsampling process in the convolutional layer and the pooling layer and does not perform an upsampling process, and

wherein the result of the process output from the output unit has a spatial dimension smaller than the size of the image input to the input unit and larger than 1×1 and represents a feature map.

2. A processing device comprising;

an input unit that receives an input of an image subject to a process;

a processing unit that subjects the image input to the input unit to a process of a convolutional neural network to which a fully connected layer is excluded; and

an output unit that outputs a result of the process in the processing unit,

wherein the convolutional neural network in the processing unit includes a convolutional layer and a pooling layer,

wherein a filter of the convolutional layer in the convolutional neural network in the processing unit is trained to learn the result of the process having a 1×1 spatial dimension,

wherein a size of the image input to the input unit is larger than a size of an image for learning input to the convolutional neural network when the filter of the convolutional layer is trained for learning, the image for learning corresponding to the result of the process used to train the taller of the convolutional layer for learning,

wherein the convolutional layer included in the convolutional neural network in the processing unit expands a border of the image by a padding and subjects an expanded image to convolution by successively shifting a filter having a size smaller than a size of the expanded image, and

wherein in the case the filter is applied to include the padding, the processing unit uses, in the padding, one of values in a portion of the image to which the filter is applied.

3. A processing device comprising:

an input unit that receives an input of an image subject to a process;

a processing unit that subjects the image input to the input unit to a process of a convolutional neural network in which a fully connected layer is excluded; and

an output unit that outputs a result of the process in the processing unit,

wherein the convolutional neural network in the processing unit Includes a convolutional layer and a pooling layer,

wherein a filter of the convolutional layer in the convolutional neural network in the processing unit is trained to learn the result of the process having a 1×1 spatial dimension,

wherein a size of the image input to the input unit is larger than a size of an image for learning input to the convolutional neural network when the filter of the convolutional layer is trained for learning, the image for learning corresponding to the result of the process used to train the filter of the convolutional layer for learning,

wherein the convolutional layer included in the convolutional neural network in the processing unit expands a border of the image by a padding and subjects an expanded image to convolution by successively shifting a filter having a size smaller than a size of the expanded image, and

wherein in the case the filter is applied to include the padding, the processing unit uses, in the padding, a statistical value of values in a portion of the image to which the filter is applied.

4. A processing method comprising:

receiving an input of an image subject to a process;

subjecting the input image to a process of a convolutional neural network in which a folly connected layer is excluded; and

outputting a result of the process,

wherein the convolutional neural network includes a convolutional layer and a pooling layer,

wherein a filter of the convolutional layer in the convolutional neural network is trained to learn the result of the process having a 1×1 spatial dimension,

wherein a size of the input image is larger than a size of an image for learning input to the convolutional neural network when the filter of the convolutional layer is trained for learning, the image for learning corresponding to the result of the process used to train the filter of the convolutional layer for learning,

wherein the convolutional neural network performs a downsampling process in the convolutional layer and the pooling layer and does not perform an upsampling process, and

wherein the result of the output process output has a spacial dimension smaller than the size of the Input image and, lamer than 1×1 and represents a feature map.

5. A processing method comprising:

receiving an input of an image subject to a process;

subjecting the input image to a process of a convolutional neural network in which a fully connected layer is excluded; and

outputting a result of the process,

wherein the convolutional neural network includes a convolutional layer and a pooling layer,

wherein a filter of the convolutional layer in the convolutional neural network is trained to learn the result of the process having a 1×1 spatial dimension,

wherein a size of the input image is larger than a size of an image for learning input to the convolutional neural network when the filter of the convolutional layer is trained for learning, the image for learning corresponding to the result of the process used to train the filter of the convolutional layer for learning,

wherein the convolutional layer included in the convolutional neural network expands a border of the Image by a padding and subjects an expanded image to convolution by successively shifting a filter having a size smaller than a size of the expanded image, and

wherein in the case the filter is applied to include the padding, one of values in a portion of the image to which the filter is applied is used in the padding.

6. A processing method comprising:

receiving an input of an image subject to a process;

subjecting the input image to a process of a convolutional neural network in which a fully connected layer is excluded; and

outputting a result of the process,

wherein the convolutional neural network includes a convolutional layer and a pooling layer,

wherein a filter of the convolutional layer in the convolutional neural network is trained to learn the result of the process having a 1×1 spatial dimension,

wherein a size of the input image is larger than a size of an image for learning input to the convolutional neural network when the filter of the convolutional layer is trained for learning, the image for learning corresponding to the result of the process used to train the filter of the convolutional layer for learning,

wherein the convolutional layer included in the convolutional neural network expands a border of the image by a padding and subjects an expanded image to convolution by successively shilling a filter having a size smaller than a size of the expanded image, and

wherein in the case the filter is applied to include the padding, a statistical value of values in a portion of the image to which the filter is applied is used in the padding.