PROCESSING METHOD AND PROCESSING DEVICE USING THIS

WIPO Patent Application WO/2019/159324

A1

An input unit 132 inputs an image to be processed. A processing unit 114 subjects the image inputted to the image unit 132 to processing with a convolutional neural network from which fully connected layers have been removed. The convolutional neural network in said processing unit 114 comprises a convolution layer and a pooling layer. An output unit 136 outputs the processing results of the processing unit 114. Here, in the convolutional neural network in the processing unit 114, the convolutional layer filter is trained with the processing results having a 1×1 spatial dimension.

JP7462738	In-vehicle cluster tracking system
JP4486780	Non-linear quantization and similarity matching method for image information retrieval
WO/2016/042037	BIAS CORRECTION IN IMAGES

HORII TOSHIHIDE

PCT/JP2018/005494

August 22, 2019

February 16, 2018

Click for automatic bibliography generation  

PANASONIC IP MAN CO LTD (JP)

G06T7/00; G06N3/08

LIN, M. ET AL.: "Network In Network", ARXIV.ORG, vol. 3, 4 March 2014 (2014-03-04), pages 1 - 10, XP055632227
KRIZHEVSKY, A. ET AL.: "Learning multiple layers of features from tiny images", 2009, pages FP - 58, XP055632229
KIMURA, TAKATO ET AL.: "Automatic Crack Detection in Infrastructure Inspection Using Fully Convolutional Network (KIMURA, Takato, CLAUS, Aranha, SAKURAI, Tetsuya. Automatic Crack Detection in Infrastructure Inspection Using Fully Convolutional Network", FULLY CONVOLUTIONAL NETWORK (MPS, vol. 2017 -MP, no. 2, 18 September 2017 (2017-09-18), pages 1 - 6, XP055723550, Retrieved from the Internet

MORISHITA Sakaki (JP)

View/Download PDF      PDF Help

Previous Patent: POWER DIVIDER/COMBINER CIRCUIT AND POWER AMPLIFICATION MODULE

Next Patent: LIQUID CRYSTAL DISPLAY PANEL