Image processing device, attack coutermeasure method, and computer readable medium
AssigneeMitsubishi Electric Corporation
InventorYoshihiro Koseki
Life of this patent
Estimating the term…
Abstract
A first detection unit (121) executes object detection for a subject image. A processing unit (130) generates a painted-out image per bounding box of the subject image by painting out the bounding box of the subject image. A second detection unit (122) executes, per painted-out image, object detection for the painted-out image. A determination unit (140) determines whether an adversarial example patch attack has been conducted, on a basis of a score value of each bounding box of the subject image and a score value of each bounding box of a painted-out image group.
Claims8
An image processing device comprising
- processing circuitry
- to execute object detection for a subject image so as to calculate a bounding box and a score value per object detected from the subject image,
- to generate a painted-out image per bounding box of the subject image by painting out the bounding box of the subject image, so as to obtain a painted-out image group,
- to execute, per painted-out image, object detection for the painted-out image, so as to calculate a bounding box and a score value per object detected from the painted-out image, and
- to determine whether an adversarial example patch attack of placing an adversarial example patch on the subject image has been conducted, on a basis of the score value of each bounding box of the subject image and the score value of each bounding box of the painted-out image group.
An attack countermeasure method comprising:
- executing object detection for a subject image so as to calculate a bounding box and a score value per object detected from the subject image;
- generating a painted-out image per bounding box of the subject image by painting out the bounding box of the subject image, so as to obtain a painted-out image group;
- executing, per painted-out image, object detection for the painted-out image, so as to calculate a bounding box and a score value per object detected from the painted-out image;
- and
- determining whether an adversarial example patch attack of placing an adversarial example patch on the subject image has been conducted, on a basis of the score value of each bounding box of the subject image and the score value of each bounding box of the painted-out image group.
A non-transitory computer readable medium recorded with an attack countermeasure program which causes a computer to execute:
- a first detection process of executing object detection for a subject image so as to calculate a bounding box and a score value per object detected from the subject image;
- a processing process of generating a painted-out image per bounding box of the subject image by painting out the bounding box of the subject image, so as to obtain a painted-out image group;
- a second detection process of executing, per painted-out image, object detection for the painted-out image, so as to calculate a bounding box and a score value per object detected from the painted-out image;
- and
- a determination process of determining whether an adversarial example patch attack of placing an adversarial example patch on the subject image has been conducted, on a basis of the score value of each bounding box of the subject image and the score value of each bounding box of the painted-out image group.
Dependencies are inferred from the claim text.