Image processing device, processing method thereof, and image processing system including the image processing device
An image processing method performed by an image processing device may include receiving input image data from an image sensor, selecting a convolution filter corresponding to each unit region from among a plurality of convolution filters based on color pattern information of the input image data, and generating, based on the selected convolution filter, a first image to be displayed from the input image data.
1 . An image processing method performed by an image processing device, the image processing method comprising:
receiving first input image data from a first image sensor;
selecting, based on color pattern information of the first input image data, a convolution filter corresponding to each unit region, from among a plurality of convolution filters; and
generating a first image to be displayed from the first input image data, based on the selected convolution filter,
wherein the first input image data includes a first pixel pattern and a second pixel pattern that is different from the first pixel pattern,
wherein the selecting comprises selecting, using a multiplexer, one of a first convolution filter and a second convolution filter based on the color pattern information of the first input image data, and applying the selected convolution filter to respective unit regions of the first pixel pattern and the second pixel pattern,
wherein each of the first convolution filter and the second convolution filter comprises a plurality of weight parameters,
wherein the first convolution filter comprises first weight parameters corresponding to the first pixel pattern,
wherein the second convolution filter comprises second weight parameters corresponding to the second pixel pattern, and
wherein the first weight parameters and the second weight parameters are different.
2 . The image processing method of claim 1 , wherein the first input image data further includes a third pixel pattern and a fourth pixel pattern, and
wherein the first to the fourth pixel patterns have R, G, and B Bayer patterns.
3 . The image processing method of claim 2 , wherein the convolution filter includes the first convolution filter, the second convolution filter, a third convolution filter, and a fourth convolution filter, respectively corresponding to the first to the fourth pixel patterns.
4 . The image processing method of claim 1 , further comprising:
training a neural network model to predict the first image to be displayed from learning data including second input image data received from a second image sensor.
5 . The image processing method of claim 4 , wherein the neural network model includes a convolution filter to which different weight parameters for each pixel are applied based on a pattern of the second input image data and phase information.
6 . The image processing method of claim 1 , wherein the image processing device includes a neural network model constructed based on at least one of a convolutional neural network-based Resnet architecture or a Unet architecture for extracting characteristics of an image by using both low-dimensional information and high-dimensional information.
7 . The image processing method of claim 1 , wherein the generating the first image comprises image processing including at least one of demosaicking for restoring a color pattern image received from the first image sensor into an RGB image to be displayed, remosaicking for changing a first color pattern image into a second color pattern image, deblurring for removing a blur from the color pattern image, or denoising for removing at least one of an artifact or a noise generated during the demosaicking or included in the color pattern image, and
wherein the color pattern image includes at least one of a Bayer pattern image, a tetra pattern image, a tetra square pattern image, a nona pattern image, or an RGBW pattern image.
8 . The image processing method of claim 1 , wherein the first input image data includes image data of at least one of color filters including a Bayer pattern, a Bayer pattern-based tetracell, a Bayer pattern-based nonacell, an RGBW pattern, and a tetra pattern.
9 . The image processing method of claim 1 , wherein the generating the first image includes performing a series of convolutions, and
wherein at least one of the series of convolutions is based on the selected convolution filter.
10 . An image processing device comprising:
a memory configured to store a program; and
a processor configured to execute the program stored in the memory and operate as instructed by the program to:
receive input image data from an image sensor and select, based on color pattern information of the input image data, a convolution filter corresponding to each unit region, from among a plurality of convolution filters, to be applied to the input image data,
wherein the input image data includes a first pixel pattern and a second pixel pattern that is different from the first pixel pattern,
wherein the processor is configured to select, using a multiplexer, one of a first convolution filter and a second convolution filter based on the color pattern information of the input image data, and apply the selected convolution filter to respective unit regions of the first pixel pattern and the second pixel pattern,
wherein each of the first convolution filter and the second convolution filter comprises a plurality of weight parameters,
wherein the first convolution filter comprises first weight parameters corresponding to the first pixel pattern,
wherein the second convolution filter comprises second weight parameters corresponding to the second pixel pattern, and
wherein the first weight parameters and the second weight parameters are different.
11 . The image processing device of claim 10 , wherein the input image data further includes a third pixel pattern and a fourth pixel pattern,
wherein the first to the fourth pixel patterns have R, G, and B Bayer patterns, and
wherein the plurality of convolution filters include the first convolution filter, the second convolution filter, a third convolution filter, and a fourth convolution filter, respectively corresponding to the first to the fourth pixel patterns.
12 . The image processing device of claim 10 , further comprising:
a neural network model constructed based on at least one of a convolutional neural network-based Resnet architecture or a Unet architecture for extracting characteristics of an image by using both low-dimensional information and high-dimensional information.
13 . The image processing device of claim 10 , wherein the processor is further configured to perform image processing including at least one of demosaicking for restoring a color pattern image received from the image sensor into an RGB image to be displayed, remosaicking for changing a first color pattern image into a second color pattern image, deblurring for removing a blur from the color pattern image, or denoising for removing at least one of an artifact or a noise generated during the demosaicking or included in the color pattern image, and
wherein the color pattern image includes at least one of a Bayer pattern image, a tetra pattern image, a tetra square pattern image, a nona pattern image, or an RGBW pattern image.
14 . The image processing device of claim 10 , wherein the input image data includes image data of at least one of color filters including a Bayer pattern, a Bayer pattern-based tetracell, a Bayer pattern-based nonacell, an RGBW pattern, and a tetra pattern.
15 . The image processing device of claim 10 , wherein the processor is further configured to generate, based on the selected convolution filter, from the input image data a first image to be displayed.
16 . An image processing system comprising:
a first image sensor configured to generate a Bayer pattern image based on a received optical signal;
an image signal processor configured to convert the Bayer pattern image received from the first image sensor into an RGB image to be displayed; and
a display configured to display the RGB image,
wherein the image signal processor is further configured to select a convolution filter corresponding to each unit region, from among a plurality of convolution filters, based on color pattern information of the Bayer pattern image,
wherein the Bayer pattern image includes a first pixel pattern and a second pixel pattern that is different from the first pixel pattern,
wherein the selecting comprises selecting, using a multiplexer, one of a first convolution filter and a second convolution filter based on the color pattern information of the Bayer pattern image, and applying the selected convolution filter to respective unit regions of the first pixel pattern and the second pixel pattern,
wherein each of the first convolution filter and the second convolution filter comprises a plurality of weight parameters,
wherein the first convolution filter comprises first weight parameters corresponding to the first pixel pattern,
wherein the second convolution filter comprises second weight parameters corresponding to the second pixel pattern, and
wherein the first weight parameters and the second weight parameters are different.
17 . The image processing system of claim 16 , wherein the Bayer pattern image further includes a third pixel pattern and a fourth pixel pattern,
wherein the first to the fourth pixel patterns have R, G, and B Bayer patterns, and
wherein the plurality of convolution filters include the first convolution filter, the second convolution filter, a third convolution filter, and a fourth convolution filter, respectively corresponding to the first to the fourth pixel patterns.
18 . The image processing system of claim 16 , wherein the image signal processor is further configured to generate the RGB image to be displayed from the Bayer pattern image based on the selected convolution filter.
19 . The image processing system of claim 16 , wherein the image signal processor is further configured to perform image processing including at least one of demosaicking for restoring a color pattern image received from the first image sensor into the RGB image to be displayed, remosaicking for changing a first color pattern image into a second color pattern image, deblurring for removing a blur from the color pattern image, or denoising for removing at least one of an artifact or a noise generated during the demosaicking or included in the color pattern image.
20 . The image processing system of claim 16 , further comprising:
a neural network device configured to train a neural network model to predict a first image to be displayed from learning data including input image data received from a second image sensor,
wherein the neural network model includes a convolution filter to which different weight parameters for each pixel are applied based on a pattern of the input image data and phase information.