Image compression device

Patent metadata

Field	Value
Publication number	US-10319115-B2
Application number	US-201715698499-A
Country	US
Kind code	B2
Filing date	Sep 7, 2017
Priority date	Mar 14, 2017
Publication date	Jun 11, 2019
Grant date	Jun 11, 2019

Abstract

Provided is an image compression device including an object extracting unit configured to perform convolution neural network (CNN) training and identify an object from an image received externally, a parameter adjusting unit configured to adjust a quantization parameter of a region in which the identified object is included in the image on the basis of the identified object, and an image compression unit configured to compress the image on the basis of the adjusted quantization parameter.

First claim

What is claimed is:

1. An image compression device comprising:
an object extracting circuit configured to perform convolution neural network (CNN) training to identify an object from an image received externally;
a parameter adjusting circuit configured to adjust a quantization parameter of a region in which the identified object is included in the image on the basis of the identified object; and
an image compression circuit configured to compress the image on the basis of the adjusted quantization parameter,
wherein the object extracting circuit sets modified-regions in the received image and performs the CNN training on a portion of the set modified-regions to identify a type of an object included in the portion of the set modified-regions,
the object extracting circuit classifies each of the modified-regions into one of a background region and an object region on a basis of pixel change amounts of the respective set modified-regions, and
wherein the object extracting circuit classifies the modified-region into the background region, when the pixel change amount is smaller than a reference value in each of the set modified-regions and classifies the modified-region into the object region when the pixel change amount is equal to or greater than the reference value.

2. The image compression device of claim 1 , wherein the object extracting circuit converts the modified-region that is classified into the object region into a warped-region and performs the CNN training using the warped-region as an input to identify a type of the object included in the object region.

3. The image compression device of claim 1 , wherein the parameter adjusting circuit sets a quantization parameter for the background region to a first value and sets a quantization parameter for the object region to a second value smaller than the first value.

4. The image compression device of claim 1 , wherein the parameter adjusting circuit sets a quantization parameter for the background area to a first value, sets a quantization parameter for the object region to a second value smaller than the first value when the type of the object included in the object region is a text, and sets the quantization parameter for the object region to a third value smaller than the second value when the type of the object included in the object region is not the text.

5. An image compression device comprising:
an object extracting circuit configured to perform convolution neural network (CNN) training to identify an object from an image received externally;
a parameter adjusting circuit configured to adjust a quantization parameter of a region in which the identified object is included in the image on the basis of the identified object; and
an image compression circuit configured to compress the image on the basis of the adjusted quantization parameter,
wherein the object extracting circuit sets modified-regions in the received image and performs the CNN training on a portion of the set modified-regions to identify a type of an object included in the portion of the set modified-regions,
wherein the object extracting circuit classifies each of the modified-regions into one of a background region and an object region on a basis of pixel change amounts of the respective set modified-regions, and
wherein the parameter adjusting circuit sets a quantization parameter for the background area to a first value, sets a quantization parameter for the object region to a second value smaller than the first value when the type of the object included in the object region is a text, and sets the quantization parameter for the object region to a third value smaller than the second value when the type of the object included in the object region is not the text.

6. An image compression device comprising:
an object extracting circuit configured to set modified-regions in an image received externally, classify each of the set modified-regions into a background region or an object region based on pixel change amount of the each of the set modified-regions, and perform convolution neural network (CNN) training on the object region among of the set modified-regions to identify a type of an object included in the object region;
a parameter adjusting circuit configured to adjust a quantization parameter of each of the set modified regions based on whether each of the set modified regions is classified into the background region or the object region, and the type of the object identified by the CNN training; and
an image compression circuit configured to compress the received image on the basis of the adjusted quantization parameter,
wherein pixel change amount of the object region is equal to or greater than a reference value, pixel change amount of the background region is smaller than the reference value.

7. The image compression device of claim 6 , wherein the object extracting circuit converts the object region into a warped-region and performs the CNN training using the warped-region as an input to identify a type of the object included in the object region.

8. The image compression device of claim 6 , wherein the parameter adjusting circuit adjusts a quantization parameter for the background region to a first value and adjusts a quantization parameter for the object region to a second value smaller than the first value.

9. The image compression device of claim 6 , wherein the parameter adjusting circuit adjusts a quantization parameter for the background area to a first value, adjusts a quantization parameter for the object region to a second value smaller than the first value when the type of the object included in the object region is a text, and adjusts the quantization parameter for the object region to a third value smaller than the second value when the type of the object included in the object region is not the text.

Assignees

• Electronics & Telecommunications Res Inst
• Electronics & Telecommunications Res Institute

Inventors

• Park Seong Mo
• Kim Sung Eun
• Kim Ju-Yeob
• Kim Jin Kyu
• Oh Kwang Il
• Lee Joo Hyun

Classifications

• G06N3/045

  Combinations of networks · CPC title

• H04N19/17

  the unit being an image region, e.g. an object · CPC title

• G06T2200/28

  involving image processing hardware · CPC title

• H04N19/124

  Quantisation · CPC title

• G06N3/08

  Learning methods · CPC title