WO2020251337A1 - 카메라 장치 및 카메라 장치의 이미지 생성 방법 - Google Patents
카메라 장치 및 카메라 장치의 이미지 생성 방법 Download PDFInfo
- Publication number
- WO2020251337A1 WO2020251337A1 PCT/KR2020/007738 KR2020007738W WO2020251337A1 WO 2020251337 A1 WO2020251337 A1 WO 2020251337A1 KR 2020007738 W KR2020007738 W KR 2020007738W WO 2020251337 A1 WO2020251337 A1 WO 2020251337A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- image
- resolution
- bayer
- deep learning
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims description 134
- 238000012549 training Methods 0.000 claims description 34
- 238000013527 convolutional neural network Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 description 186
- 238000013135 deep learning Methods 0.000 description 145
- 239000000872 buffer Substances 0.000 description 68
- 230000008569 process Effects 0.000 description 68
- 238000004422 calculation algorithm Methods 0.000 description 60
- 238000010586 diagram Methods 0.000 description 33
- 230000006870 function Effects 0.000 description 32
- 238000007781 pre-processing Methods 0.000 description 30
- 238000005516 engineering process Methods 0.000 description 27
- 230000015654 memory Effects 0.000 description 27
- 238000005070 sampling Methods 0.000 description 17
- 230000000875 corresponding effect Effects 0.000 description 16
- 230000005540 biological transmission Effects 0.000 description 15
- 238000004566 IR spectroscopy Methods 0.000 description 14
- 238000013528 artificial neural network Methods 0.000 description 14
- 238000004891 communication Methods 0.000 description 14
- 230000003287 optical effect Effects 0.000 description 14
- 230000008901 benefit Effects 0.000 description 12
- 230000000694 effects Effects 0.000 description 11
- 238000013136 deep learning model Methods 0.000 description 9
- 238000013461 design Methods 0.000 description 9
- 239000007788 liquid Substances 0.000 description 8
- 238000003672 processing method Methods 0.000 description 8
- 239000003086 colorant Substances 0.000 description 7
- 230000009471 action Effects 0.000 description 6
- 238000012937 correction Methods 0.000 description 6
- 239000004065 semiconductor Substances 0.000 description 6
- 230000015556 catabolic process Effects 0.000 description 5
- 238000006731 degradation reaction Methods 0.000 description 5
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000011664 signaling Effects 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 230000033001 locomotion Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 241001465754 Metazoa Species 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 239000000470 constituent Substances 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 239000011521 glass Substances 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 229910044991 metal oxide Inorganic materials 0.000 description 2
- 150000004706 metal oxides Chemical class 0.000 description 2
- 238000002203 pretreatment Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000011426 transformation method Methods 0.000 description 2
- 239000012780 transparent material Substances 0.000 description 2
- 210000000857 visual cortex Anatomy 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000002542 deteriorative effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 229920001690 polydopamine Polymers 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000003381 stabilizer Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/80—Camera processing pipelines; Components thereof
- H04N23/815—Camera processing pipelines; Components thereof for controlling the resolution by using a single image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/10—Circuitry of solid-state image sensors [SSIS]; Control thereof for transforming different wavelengths into image signals
- H04N25/11—Arrangement of colour filter arrays [CFA]; Filter mosaics
- H04N25/13—Arrangement of colour filter arrays [CFA]; Filter mosaics characterised by the spectral characteristics of the filter elements
- H04N25/134—Arrangement of colour filter arrays [CFA]; Filter mosaics characterised by the spectral characteristics of the filter elements based on three different wavelength filter elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Definitions
- the present invention relates to a camera module, a mobile terminal device including the same, and a method of generating an image, and more particularly, to a technology for generating a low-resolution image as a high-resolution image using a deep learning algorithm.
- Such a camera module is manufactured with an image sensor such as a CCD or CMOS as a main component, and is manufactured so that focus can be adjusted to adjust the size of an image.
- an image sensor such as a CCD or CMOS
- Such a camera module includes a plurality of lenses and an actuator, and the actuator moves each lens to change the relative distance, so that the optical focal length can be adjusted.
- the camera module includes an image sensor that converts an externally received optical signal into an electrical signal, a lens that condenses light with the image sensor, an IR (Infrared) filter, a housing including these, and a printing that processes signals from the image sensor.
- a circuit board and the like are included, and the focal length of the lens is adjusted by an actuator such as a VCM (Voice Coil Motor) actuator or a MEMS (Micro Electromechanical Systems) actuator.
- VCM Vehicle Coil Motor
- MEMS Micro Electromechanical Systems
- a camera is equipped with a zoom function to capture a distant object.
- the zoom function is an optical zoom that enlarges the object by moving the actual lens inside the camera, and a partial screen of image data photographing the object. It is divided into a digital zoom method that obtains a zoom effect by displaying enlarged in a digital processing method.
- sensor shift technology that shakes the sensor with Voice Coil Motor (VCM) or Micro-Electro Mechanical Systems (MEMS) technology
- MEMS Micro-Electro Mechanical Systems
- OIS Optical Image
- technologies for realizing high-resolution images by generating more pixel information by moving parts inside the camera such as a stabilizer technology and a technology that shakes a filter between a sensor and a lens.
- the size of the camera module increases as the camera module is inserted in a complex device for implementing this, and it is difficult to use in a vehicle with a camera, and can only be used in a fixed environment because it is implemented by shaking the parts.
- the present invention is an invention devised to solve the problems of the prior art as described above, and can generate a high-resolution image without causing problems such as motion blur or artifacts. It is to provide a camera module and a mobile terminal device including the same.
- a chip to which a high-resolution implementation algorithm based on deep learning technology is applied is mounted on a camera module or a mobile terminal device equipped with a camera module in the form of On The Fly to more efficiently generate high-resolution images. It is to provide a capable camera module and a mobile terminal device including the same.
- the camera device includes an image sensor that generates first Bayer data having a first resolution and a second Bayer having a second resolution higher than the first resolution by performing deep learning based on the first Bayer data. It may include a processor that outputs data.
- the processor may generate first array data in which the first Bayer data are arranged for each wavelength band, and then perform deep learning based on the generated first array data to generate the second array data.
- the processor may generate the second Bayer data based on the second array data.
- the processor may generate an image having the second resolution based on the second Bayer data.
- the processor includes a first data alignment unit that generates first array data by arranging the first Bayer data by wavelength band, and deep learning that outputs second array data by performing deep learning based on the first array data. It may include a processor and a second data alignment unit that generates second Bayer data in which the second array data is arranged in a Bayer pattern.
- the processor is configured to generate at least one first line buffer for storing the first Bayer data for each line, and first array data arranged for each wavelength band by receiving information output from the first line buffer.
- a first data alignment unit, a deep learning processor that performs deep learning based on the first array data to generate second array data, and second data that generates second array data that arranges second array data in a Bayer pattern It may include an alignment unit and at least one second line buffer for storing data output from the second data alignment unit for each line.
- a method of generating an image of a camera device includes generating first Bayer data having a first resolution, generating first array data by classifying the first Bayer data by wavelength band, and the first array Generating second array data by performing deep learning based on data, and generating second Bayer data having a second resolution higher than the first resolution based on the second array data.
- the generating of the second Bayer data may include generating the second array data by arranging them in a Bayer pattern.
- the first Bayer data includes a plurality of row data
- the generating of the first array data includes the first Bayer data output through a preset N+1 row line. It may include the step of generating the first arrangement data based on.
- the outputting through the preset N+1 row lines includes sequentially storing N row data among a plurality of row data of the received first Bayer data, and then transmitting the N+1 row. It may include the step of outputting N row data together.
- the chip to which the present technology is applied can be manufactured in a small size, it can be mounted in various ways in various positions according to the purpose of use of the mounted device, so that the degree of freedom of design can be increased.
- an expensive processor is not required to perform the existing deep learning algorithm, a high-resolution image can be generated more economically.
- this technology can be implemented in a manner that can be mounted at any position of an image sensor module, camera module, or AP module, the technology is applied to a camera module without zoom function or a camera module that supports only fixed zoom for a specific magnification You can use the zoom function.
- FIG. 1 is a block diagram showing some components of a camera module according to a first embodiment of the present invention.
- FIG. 2 is a view showing some components of the image generating unit according to the first embodiment of the present invention.
- FIG. 3 is a diagram illustrating a process of performing deep learning training according to the first embodiment of the present invention.
- FIG. 4 is a diagram showing a process of performing deep running training according to the first embodiment and another embodiment of the present invention.
- FIG. 5 is a diagram illustrating information input to a processor to which a deep learning algorithm is applied and output information output through the processor.
- FIGS. 6 and 7 are diagrams illustrating a state in which a first Bayer image having a first resolution is converted into a second Bayer image having a second resolution by a processor.
- FIG. 8 is a block diagram showing some components of a mobile terminal device according to the first embodiment of the present invention.
- FIG. 9 is a block diagram showing some components of a mobile terminal device including a camera module according to the first embodiment of the present invention.
- FIG. 10 is a block diagram showing some components of a mobile terminal device including a camera module according to another embodiment of the first embodiment of the present invention.
- FIG. 11 is a block diagram showing some components of a mobile terminal device including a camera module according to another embodiment of the first embodiment of the present invention.
- FIG. 12 is a flow chart showing the procedure of an image generating method according to the first embodiment of the present invention.
- FIG. 13 is a block diagram of an image processing apparatus according to a second embodiment of the present invention.
- FIG. 14 is a block diagram of a camera module included in the image processing apparatus according to the second embodiment of the present invention.
- 15 is a block diagram of a mobile device to which the configuration of an image processing device according to a second embodiment of the present invention is applied.
- 16 is a block diagram of an image processing apparatus according to another embodiment of the second embodiment of the present invention.
- 17 is a block diagram of a mobile device to which the configuration of an image processing apparatus according to another embodiment of the second embodiment of the present invention is applied.
- FIG. 18 is a diagram for explaining a process of processing an image in the image processing apparatus according to the second embodiment of the present invention.
- 19 is a block diagram of an image processing apparatus according to another embodiment of the second embodiment of the present invention.
- FIG. 20 is a flowchart of an image processing method according to a second embodiment of the present invention.
- 21 is a flowchart of an image processing method according to another embodiment of the second embodiment of the present invention.
- FIGS. 13 to 21 are an image processing device and image processing according to a second embodiment of the present invention. It is a diagram of the method.
- FIG. 1 is a block diagram showing some components of the camera module 100 according to the first embodiment of the present invention
- FIG. 2 is a block diagram showing some components of the image generating unit 200 according to the first embodiment
- 3 is a diagram illustrating an example of a deep learning process performed by the processor 220 according to the first embodiment.
- the camera module 100 transmits an image sensor 110 that acquires an image of an external object and an image acquired by the image sensor 110 to the image generator 200.
- the image generation unit 200 receives the image transmitted by the transmission unit 120 and transmits the image to the processor 220, the receiving unit 210, the image received from the receiving unit 210
- a processor 220 that generates an image having a resolution different from the resolution of the received image by applying an algorithm derived from the deep learning training, and an output unit that receives and transmits the image generated by the processor 220 to an external device (230) and the like.
- the algorithm derived as a result of performing the deep learning training may be the convolutional neural network described above.
- the processor 220 may be a processor learned using a deep learning-based algorithm. It may include a pipelined processor, and may include a convolutional neural network trained to generate second Bayer data from first Bayer data.
- the image sensor 110 may include an image sensor such as a Complementary Metal Oxide Semiconductor (CMOS) or a Charge Coupled Device (CCD) that changes light entering through the lens 120 of the camera module into an electric signal.
- CMOS Complementary Metal Oxide Semiconductor
- CCD Charge Coupled Device
- the transmission unit 120 may transmit the image acquired by the image sensor 110 to the reception unit 210 of the image generating apparatus 200.
- the transmission unit 120 and the image sensor 110 are illustrated as distinct components, but the present invention is not limited thereto, and the image sensor 110 may simultaneously perform the role of the transmission unit 120 to be described later.
- the transmission unit 120 may extract information on a Bayer pattern from an image acquired by the image sensor 110 and transmit the information about the Bayer pattern to the reception unit 210.
- the image generation unit 200 receives the image transmitted by the transmission unit 120 and transmits the image to the processor 220 by the transmission unit 210, the image received from the transmission unit 210 by deep learning training.
- a processor 220 that generates an image having a higher resolution using the generated algorithm, and an output unit 230 that receives and transmits the image generated by the processor 220 to the outside may be included.
- the processor 220 receives a Bayer image having a first resolution from the receiving unit 210, and then generates a Bayer image having a second resolution using an algorithm generated by deep learning training, and then generates The second Bayer image can be transmitted to the output unit 230.
- the second resolution means a resolution having a resolution value different from the first resolution, and specifically, means a resolution higher or lower than the first resolution. can do.
- the resolution value that the second resolution can have can be freely set and changed by the user according to the user's purpose.
- the camera module 100 may further include an input unit for receiving information on the second resolution, although not shown in the drawing, and the user can transmit information on the desired resolution to the camera module 100. You can enter as
- the second resolution can be set to a resolution that has a large difference from the first resolution, and when a new image is to be acquired within a relatively short time, the first resolution and The second resolution value can be freely set to a resolution that does not have much difference in resolution.
- the processor 220 may be implemented through a memory (not shown) in which at least one program command executed through the processor is stored.
- the memory may include a volatile memory such as S-RAM and D-lap.
- the present invention is not limited thereto, and in some cases, the memory is a ratio of flash memory, ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory: EPROM), and EPROM (Electrically Erasable Programmable Read Only Memory: EEPROM). It may also include volatile memory.
- a typical camera device or camera module outputs image data through a process (color interpolation, color interpolation, or demosaicing) that receives Bayer patterns from an image sensor and coats them with color, and the transmission unit 120 according to an embodiment May extract information including Bayer pattern information from the image received from the image sensor 110 and transmit the extracted information to the outside.
- a process color interpolation, color interpolation, or demosaicing
- the transmission unit 120 May extract information including Bayer pattern information from the image received from the image sensor 110 and transmit the extracted information to the outside.
- the Bayer pattern may include raw data output from the image sensor 110 that converts an optical signal included in the camera device or camera module 100 into an electrical signal.
- the optical signal transmitted through the lens 120 included in the camera module 100 is converted into an electrical signal through each pixel disposed on an image sensor capable of detecting R, G, and B colors. Can be.
- the specification of the camera module 100 is 5 million pixels, it may be considered that an image sensor including 5 million pixels capable of detecting R, G, and B colors is included. Although the number of pixels is 5 million, a monochromatic pixel that does not actually detect each color but only detects the brightness of black and white can be viewed in a form combined with any one of R, G, and B filters.
- R, G, and B color filters are arranged in a specific pattern on monochromatic pixel cells arranged by the number of pixels. Accordingly, R, G, and B color patterns are intersected and arranged according to the visual characteristics of the user (ie, human), which is called a Bayer pattern.
- the Bayer pattern has a smaller amount of data than image data. Therefore, even a device equipped with a camera module that does not have a high-end processor can transmit and receive Bayer pattern image information relatively faster than image data, and convert it into images having various resolutions based on this. There are advantages.
- a camera module is mounted on a vehicle, so that many processors are not required for image processing even in an environment in which the camera module uses a low voltage differential signaling method (LVDS) with a full-duplex transmission rate of 100 Mbit/s. As there is no overload, it may not be a hazard to the driver or the driver's safety.
- LVDS low voltage differential signaling method
- the transmission unit 120 receives the Bayer pattern frame from the image sensor 110, and then down-samples it to a size of 1/n. Can be sent.
- the transmission unit 120 may perform downsampling after performing smoothing through a Gaussian filter or the like on data of the Bayer pattern received before downsampling. Thereafter, a frame packet may be generated based on the down-sampled image data, and then the completed frame packet may be transmitted to the receiver 210.
- these functions may be performed at the same time in the processor 220 instead of the transmission unit 120.
- the transmission unit 120 may include a serializer (not shown) that converts Bayer patterns into serial data in order to transmit Bayer pattern information in a serial communication method such as a low voltage differential signaling method (LVDS).
- a serializer (not shown) that converts Bayer patterns into serial data in order to transmit Bayer pattern information in a serial communication method such as a low voltage differential signaling method (LVDS).
- LVDS low voltage differential signaling method
- the serializer may include or be implemented with a buffer that temporarily stores data and a phase-locked loop (PLL) that forms a period of transmitted data.
- PLL phase-locked loop
- the algorithm applied to the processor 220 of the camera module 100 is an algorithm that generates an image having a resolution higher than the resolution of the input image, and is an optimum generated by repeatedly performing deep learning training. Can mean the algorithm of.
- the convolutional neural network which is an algorithm generated by deep learning training, may be trained to receive first Bayer data having one resolution and generate second Bayer data having a second resolution.
- Deep learning sometimes expressed as deep learning, is machine learning that attempts to achieve a high level of abstraction (summarizing key contents or functions in a large amount of data or complex data) through a combination of several nonlinear transformation methods. It refers to a set of algorithms for (machine learning).
- deep learning represents some training data in a form that a computer can understand (for example, in the case of an image, pixel information is expressed as a column vector), and there are many to apply it to learning.
- a learning technique for research how to make a better representation technique and how to make a model to learn these
- learning techniques such as Deep Neural Networks (DNN) and Deep Belief Networks (DBN).
- deep learning may first recognize the surrounding environment and transmit the current environment state to the processor.
- the processor performs the corresponding action, and the environment again informs the processor of the reward for the action. And the processor chooses the action that maximizes the reward. Through this process, the learning process can be repeated.
- the training data used while performing deep learning may be a result obtained by converting a Bayer image with a low resolution into a Bayer image with a high resolution, or may be information obtained through simulation.
- FIG. 3 is a diagram illustrating a process of performing deep learning training according to an embodiment
- FIG. 4 is a diagram illustrating a process of performing deep learning training according to another embodiment.
- the deep learning of FIG. 3 is deep learning to which a deep neural network (DNN) algorithm is applied, and is a diagram illustrating a process of generating an image having a new resolution according to the application of the DNN algorithm.
- DNN deep neural network
- a deep neural network is a deep neural network in which multiple hidden layers exist between an input layer and an output layer, and a connection pattern between neurons similar to the structure of an animal's visual cortex. It can be embodied as a convolutional neural network that forms a and a recurrent neural network that builds up the neural network every moment over time.
- the convolutional neural network may be at least one model of a Fully Convolutional Network (FCN), U-Net, MobileNet, Residual Dense Network (RDN), and Residential Channel Attention Network (RCAN). It is natural that a variety of other models can be used.
- DNN classifies neural networks by reducing and distorting the amount of data by repeating convolution and sub-sampling.
- DNN outputs class results through feature extraction and classification behavior, and is mainly used to analyze images, and convolution means image filtering.
- the processor 220 attempts to increase the magnification based on the Bayer image 10 having the first resolution received from the receiver 210. Convolution and sub-sampling are performed on the desired region.
- Increasing the magnification means magnifying only a specific part of the image acquired by the image sensor 110. Accordingly, since a portion not selected by the user is a portion that the user is not interested in, there is no need to perform a process of increasing the resolution, and thus the convolution and sub-sampling process can be performed only on the portion selected by the user.
- Sub-sampling refers to the process of reducing the size of an image.
- a Max Pool method may be used for sub-sampling.
- Max-Pull is a technique that selects the maximum in a given area, similar to how neurons respond to the largest signal.
- Sub-sampling has the advantage of reducing noise and increasing the speed of learning.
- a plurality of images 20 may be output as shown in FIG. 3. Thereafter, a plurality of images having different characteristics may be output using an up-scale method based on the output images.
- the up-scale method means that the image is scaled up by r*r times by using different r ⁇ 2 filters.
- the processor 220 When a plurality of images according to the upscale are output as shown in FIG. 3 (30), the processor 220 recombines based on these images and finally outputs a second Bayer image 40 having a second resolution. can do.
- the deep learning of FIG. 4 is a diagram illustrating a method of performing deep learning in a manner different from that of the deep learning of FIG. 3, and is a diagram specifically describing a process of generating an algorithm formed by iterative learning.
- deep learning training may be performed based on the input.
- the deep learning according to FIG. 4 uses the first sample data (X) as input data and compares and analyzes the output data (Y) and the second sample data (Z) for performing deep learning training.
- An algorithm that generates an image with a higher resolution can be created as a basis.
- the output data (Y) is data actually output through deep learning
- the second sample data (Z) is data input by the user.
- the first sample data (X) is input to the algorithm, it is most ideally output. It can mean data that can be made.
- the algorithm according to FIG. 4 compares and analyzes the most ideal second sample data Z as the output result and the first output data Y, which is the actual output data, to derive the difference, and then the algorithm in the direction of canceling the difference. You can give feedback on
- feedback is given by changing or deleting the parameter or creating a new parameter to provide the ideal output data, the second sample data Z, and the actual output data, the first output. There is no difference in data (Y).
- Fig. 4 there are a total of 3 layers (L1, L2, L3) that affect the algorithm, and a total of 8 parameters (P11, P12, P13, P21, P22, P31, P32) are in each layer. I assume it exists.
- the feedback is P22.
- Algorithms can be changed in the direction of decreasing parameters.
- the feedback is P33.
- the algorithm to which deep learning is applied through this method may cause the first output data Y to be actually output to be similar to the second sample data Z, which is the most ideal output data.
- the resolution of the second sample data Z may be the same as or higher than the resolution of the first output data Y, and the resolution of the second sample data Z may be the same as the resolution of the first output data Y. It can be the same.
- the deep learning process and the number of memory gates must be minimized.
- the factors that most affect the number of gates are algorithm complexity and clock ( It is the amount of data processed per clock, and the amount of data processed by the processor depends on the input resolution.
- the processor 220 generates an image with a high magnification by reducing the input resolution in order to reduce the number of gates and then up-scaling it later, so there is an advantage of generating an image faster. do.
- zoom 2x by upscaling the width and height by 2x each based on a 1/4 area (2Mp).
- 2Mp an input resolution
- the horizontal and vertical are respectively upscaled by 4 times based on the generated image. If you zoom 4x with the (Up scailing) method, you can create a zoomed image of the same area as the 2x zoom.
- deep learning in order to prevent performance degradation due to loss of input resolution, deep learning generates an image by learning by a magnification corresponding to the resolution loss, thereby minimizing performance degradation.
- the processor 220 applies an algorithm that has already been generated through deep learning, so it can be easily applied to a low-end camera module and various devices including the same. Since high resolution is implemented by using only a line buffer, there is also an effect of implementing a processor with a relatively small chip.
- FIG. 5 is a block diagram showing some components of the processor 220 according to an embodiment.
- the processor generates a plurality of line buffers 11 for receiving first Bayer data, and first array data for arranging first Bayer data output through the line buffer for each wavelength band.
- the first data alignment unit 221 to perform deep learning according to a preset algorithm, the deep learning processor 222 to perform deep learning, and the second array data output through the deep learning processor 222 are arranged in a Bayer pattern and A second data alignment unit 223 for generating 2-bayer data and a plurality of line buffers 12 for outputting second Bayer data output through the second data alignment unit 223 may be included.
- the first Bayer data is information including the Bayer pattern described above, and is described as Bayer data in FIG. 5, but may be defined as a Bayer image or a Mayer pattern.
- first data alignment unit 221 and the second data alignment unit 223 are illustrated as separate components for convenience, but are not limited thereto, and the deep learning processor 222 A function performed by the alignment unit 221 and the second data alignment unit 223 may be performed together.
- the first Bayer data having the first resolution received by the image sensor 110 contains (n+1) line buffers 11a, 11b, and 11n for image information on an area selected by the user. .11n+1) can be transmitted.
- the Bayer image having the second resolution is generated only for the region selected by the user, image information on the region not selected by the user is not transmitted to the line buffer 11.
- the first Bayer data includes a plurality of row data
- the plurality of row data may be transmitted to the first data alignment unit 221 through the plurality of line buffers 11.
- the area in which deep learning is to be performed by the deep learning processor 222 is a 3 X 3 area
- a total of 3 lines must be simultaneously transmitted to the first data alignment unit 221 or the processor 220 to deep learning. Can be done. Accordingly, information on the first line of the three lines is transmitted to the first line buffer 11a and then stored in the first line buffer 11a, and the information on the second line of the three lines is transmitted to the second line buffer. After being transmitted to (11b), the second line buffer (11b) may be stored.
- the third line since there is no information on a line to be received thereafter, it is not stored in the line buffer 11 and may be immediately transmitted to the processor 220 or the first data alignment unit 221.
- the first data alignment unit 221 or the processor 220 needs to receive information on three lines at the same time, the first line stored in the first line buffer 11a and the second line buffer 11b is The information on the second line and the information on the second line may be simultaneously transmitted to the processor 220 or the first image alignment unit 219.
- the area in which deep learning is to be performed by the deep learning processor 222 is a (N+1) x (N+1) area
- a total of (N+1) lines are the first data alignment unit 221 or Deep learning can be performed only when it is simultaneously transmitted to the processor 220. Accordingly, information on the first line of (N+1) lines is transmitted to the first line buffer 11a, and then stored in the first line buffer 11a, and the second line of the (N+1) lines
- Information on the second line buffer 11b may be stored after being transmitted to the second line buffer 11b, and information on the Nth line among (N+1) lines is the Nth line buffer 11n. After being transmitted to, the N-th line buffer 11n may be stored.
- the first image alignment unit 219 After receiving Bayer data from the line buffer 11, the first image alignment unit 219 generates first arrangement data by arranging Bayer data by wavelength band, and then converts the generated first arrangement data into a deep learning processor 222 ) Can be sent.
- the first image aligning unit 219 may generate first array data arranged by classifying the received information into specific wavelengths or specific colors (Red, Green, Blue).
- the deep learning processor 222 may generate second array data by performing deep learning based on the first array data received through the first image alignment unit 219.
- performing deep learning may mean a process of generating an algorithm through inference or iterative learning in order to generate an optimal algorithm as described with reference to FIGS. 3 and 4 above, but at the same time, Executing the generated algorithm can also be regarded as performing deep learning.
- the deep learning processor 222 performs deep learning based on the first array data received through the first image alignment unit 219 to provide second array data having a second resolution higher than the first resolution. Can be created.
- the first array data is received for the 3 x 3 area
- deep learning is performed for the 3 x 3 area
- the first array is for the (n+1) x (n+1) area.
- deep learning can be performed on the (n+1) x (n+1) region.
- the second array data generated by the deep learning processor 222 is transmitted to the second data alignment unit 223, and the second data alignment unit 223 converts the second alignment data to a second array having a Bayer pattern. Can be converted to Bayer data.
- the converted second Bayer data is output to the outside through a plurality of line buffers 12a, and the outputted second Bayer data is converted to an image having a second resolution that is higher than the first resolution by another process. Can be created.
- FIGS. 6 and 7 are diagrams illustrating a state in which a first Bayer image having a first resolution image is converted into a second Bayer image having a second resolution by the processor 220.
- the processor 220 may perform an image conversion process for that region, and as a result of the execution, as shown in FIGS. 6 and 7 A Bayer image 40 having a second resolution may be generated.
- FIG. 8 is a block diagram showing some components of a mobile terminal device 400 according to an embodiment.
- a mobile terminal device 400 includes a filter 110, a lens 120, an image sensor 130, a transmitter 140, a driver IC 150, and an actuator. It may include an AP 300 including 160, a receiver 210, a processor 220, an output unit 230, and an ISP 310.
- the image sensor 130, the transmitting unit 140, the receiving unit 210, the processor 220, and the output unit 230 are constituent elements that have the same roles as those described in FIGS. 1 and 2, and thus description thereof will be omitted.
- the filter 110 serves to selectively block light that is introduced from the outside, and may be generally located above the lens 120.
- the lens 120 is a device that finely grinds a surface of a transparent material such as glass into a spherical surface to collect or diverge light coming from an object to form an optical image, and a general lens 120 used in the camera module 100 includes a plurality of Lenses having different characteristics may be provided.
- Driver IC 150 refers to a semiconductor (IC) that provides driving signals and data as electrical signals to the panel so that a character or video image is displayed on the screen, and as will be described later, the driver IC is the mobile terminal device 400. It can be placed in a variety of positions. In addition, the driver IC (150m Driver IC) may drive the actuator 160.
- IC semiconductor
- the driver IC 150m Driver IC
- the actuator may adjust the focus by adjusting the position of the lens or the barrel including the lens.
- the actuator 160 may be a VCM (Voice Coil Motor) type.
- the lens 120 may include a variable focus lens.
- the driver IC may drive the varifocal lens.
- the lens may include a liquid lens containing a liquid, and in this case, the driver IC may adjust the focus by adjusting the liquid in the liquid lens.
- An application processor is a mobile memory chip and refers to a core semiconductor that operates various applications and processes graphics in the mobile terminal device 400.
- the AP 300 can be implemented in the form of a System on Chip (SoC) that includes all the functions of a computer's central processing unit (CPU) and a chipset that controls the connection of other equipment such as memory, hard disk, and graphics card. have.
- SoC System on Chip
- the image signal processing unit (ISP) 300 may receive the second Bayer image generated by the processor 220 using Mobile Industry Processor Interface (MIPI) communication and perform an image signal processing process.
- MIPI Mobile Industry Processor Interface
- the image signal processing unit 300 may include a plurality of sub-processes while processing the image signal. For example, one or more of a gamma correction, color correction, auto exposure correction, and auto white balance process may be included for the received image.
- 9 to 11 are block diagrams illustrating various embodiments in which the driver IC 150 and the processor 220 may be disposed inside the mobile terminal device 400.
- FIGS. 9 to 11 Each of the constituent elements illustrated in FIGS. 9 to 11 has been described in detail above, and thus a position where the driver IC 150 and the processor 220 can be disposed will be mainly described without further description.
- driver IC 150 and the processor 220 may be independently implemented as separate modules in the camera module 100 as shown in FIG. 7.
- the processor 220 may be implemented in the form of a chip and included in the image generating unit 200, and the driver IC 150 may be implemented as a separate chip independently from the image generating unit 200 and the processor 220. I can.
- the image generating unit 200 is illustrated as including a receiving unit 210, a processor 220, and an output unit 230, respectively, but the present invention is not limited thereto, and the image generating unit 200 only includes the processor 220.
- the processor 220 may simultaneously perform the functions of the receiving unit 210 and the output unit 230 described above.
- the present technology can be applied by inserting the chip on which the image generating unit 200 is implemented into an existing camera module. Regardless of the structure of the module, there is an effect that can implement this technology.
- driver IC 150 and the processor 220 may be implemented as a single module in the image generator 200 included in the camera module 100 as shown in FIG. 8.
- the image generating unit 200 may be implemented as a single chip. In this case, the image generating unit 200 may simultaneously perform the roles of the driver IC 150 and the processor 220.
- the image generating unit 200 is illustrated as including the receiving unit 210, the processor 220, and the output unit 230, respectively, but the present invention is not limited thereto, and the image generating unit 200 includes the processor 220 and the It includes only the driver IC 150, and the processor 220 may simultaneously perform the functions of the receiving unit 210, the output unit 230, and the driver IC 150 described above.
- driver IC 150 and the processor 220 are implemented in the form as shown in FIG. 10, it is more economical because the role of the driver IC 150 and the role of the processor 220 can be simultaneously performed using one chip. There is an effect of manufacturing the camera module 100 as a result.
- the driver IC 150 and the processor 220 as shown in FIG. 9, the driver IC 150 is mounted inside the camera module 100, the processor 220 is the camera module 100 Apart from and may be disposed inside the mobile terminal device 400.
- the processor 220 may be implemented in the form of a chip and included in the image generating unit 200, and the driver IC 150 is implemented independently of the image generating unit 200 and the processor 220, and the camera module 100 ) Can be placed inside.
- the image generating unit 200 is illustrated as including the receiving unit 210, the processor 220, and the output unit 230, respectively, but the present invention is not limited thereto, and the image generating unit 200 includes only the processor 220.
- the processor 220 may simultaneously perform the functions of the receiving unit 210 and the output unit 230 described above.
- the present technology can be implemented by inserting the chip on which the image generating unit 200 is implemented into an existing camera module. Regardless of the structure of the module, there is an advantage of implementing this technology. In addition, there is an effect of reducing the thickness of the module itself compared to the high image sensor.
- FIG. 12 is a flowchart illustrating a procedure of an image generating method according to an exemplary embodiment.
- a first Bayer image having a first resolution may be received from the image sensor 110 (S110).
- information about the second resolution may be received from the user.
- information on the second resolution may be received from a user through a separate input device.
- the second Bayer image may be generated by using an algorithm generated through deep learning to generate a Bayer image having the second resolution set by the user thereafter (S130). , S140)
- the camera module, the mobile terminal device including the same, and the image generation method according to the embodiment implements high resolution by using only a few line buffers, so that it can be implemented with a relatively small chip.
- the chip to which this technology is applied can be manufactured in a small size, it can be mounted in various ways in various locations depending on the purpose of use of the mounted device, increasing the degree of design freedom, and the algorithm generated by deep learning There is an advantage in that a high-resolution image can be generated more economically because an expensive processor is not required using a worn processor.
- this technology can be implemented by mounting a simple chip on the camera module, it is possible to use the continuous zoom function by applying this technology to a camera module without a zoom function or a camera module that supports only fixed zoom for a specific magnification. have.
- the first Bayer data output using the learned convolutional neural network is input to the image signal processor.
- An RGB image may be generated by performing demosaic (RGB conversion) of an image signal processor of the AP.
- the processor for generating the second Bayer data from the first Bayer data may be implemented as a front end of the image signal processor (software logic of an AP, that is, a preprocessing logic of an ISP front end), as a separate chip, or in a camera module.
- Bayer data which is raw data, has a bit resolution of 10 bits or more, whereas in the case of RGB data that has undergone image processing by ISP, data loss such as Noise/Artifact Reduction and Compression occurs in ISP.
- RGB data is 8 bits, and the information it contains is considerably reduced.
- ISP includes nonlinear processing such as tone mapping, making it difficult to process image restoration, but Bayer data has linearity proportional to light, so image restoration can be easily processed. have.
- the signal-to-noise ratio (PSNR) is also increased by 2 to 4 dB when using the same algorithm compared to using RGB data, which is the use of Bayer data, and through this, multi-frame de-noise or SR performed in the AP It can be treated effectively. That is, by using Bayer data, high-resolution conversion performance can be improved, and since Bayer data is output, additional image processing performance of the AP can also be improved.
- FIGS. 13 to 18 an image processing apparatus and an image processing method according to a second exemplary embodiment of the present invention will be described with reference to FIGS. 13 to 18.
- Detailed description of the image processing apparatus and the image processing method according to the second embodiment of the present invention is an imaging process, a camera module, an image generating unit, an imaging device, a mobile terminal device, a camera device, and a method according to the first embodiment of the present invention.
- the image processing apparatus 1130 includes a camera module 1110 and an AP module 1120, and the AP module 1120 includes a first processing unit 1121 and a second processing unit 1122 Consists of In addition, one or more processors may be further included, or one or more memories or communication units may be further included.
- the image processing device 1130 refers to a device including a function of processing an image, and may mean any one of electronic devices, such as a mobile terminal including an image processing module to an image processing unit.
- the camera module 1110 includes an image sensor.
- the camera module 1110 outputs Bayer data of the first resolution from the image sensor.
- the camera module 1110 may include a lens 1111, an image sensor 1112, a sensor board 1113 on which an image sensor 1112 is mounted, and a connector 1114 that transmits and receives data to and from the outside. have.
- the above components may be formed as one module. That is, it is an independent device that is distinguished from components other than the camera module, and may be implemented in a form of transmitting and receiving data with other modules.
- the lens 1120 is a device that finely grinds a surface of a transparent material such as glass into a spherical surface to collect or diverge light coming from an object to form an optical image, and a general lens 1111 used in the camera module 1110 is Lenses having different characteristics may be provided. A filter may be formed on the lens 1111 to selectively block light introduced from the outside.
- the image sensor 1112 may include an image sensor such as a Complementary Metal Oxide Semiconductor (CMOS) or a Charge Coupled Device (CCD) that converts light entering through a lens of a camera module into an electrical signal.
- CMOS Complementary Metal Oxide Semiconductor
- CCD Charge Coupled Device
- the image sensor 1112 may generate Bayer data including Bayer pattern information through a color filter on the acquired image. Bayer data may have a first resolution according to a specification of the image sensor 1112 or a zoom magnification set when a corresponding image is generated.
- the image sensor 1112 may be formed on the sensor board 1113.
- the lens 1111 may also be formed on the sensor board 1113.
- the optical signal transmitted through the lens 1111 may be converted into an electric signal through each pixel disposed in the image sensor 1112 capable of detecting R, G, and B colors.
- the specification of the camera module 1110 is 5 million pixels, it can be considered that an image sensor 1112 including 5 million pixels capable of detecting R, G, and B colors is included.
- the number of pixels is 5 million, a monochromatic pixel that does not actually detect each color, but only detects the brightness of black and white, can be viewed in a form combined with any one of R, G, and B filters. That is, in the image sensor 1112, R, G, and B color filters are arranged in a specific pattern on monochrome pixel cells arranged by the number of pixels.
- R, G, and B color patterns are intersected and arranged according to the visual characteristics of the user (ie, human), which is called a Bayer pattern.
- the Bayer pattern has a smaller amount of data than image data. Therefore, even in a device equipped with the camera module 1110 that does not have a high-end processor, it is possible to transmit and receive Bayer pattern image information relatively faster than image data, and convert it into images of various resolutions based on this. There is an advantage that can be done.
- the camera module 1110 is mounted on a vehicle, so that the camera module 1110 performs image processing even within an environment in which a low voltage differential signaling method (LVDS) having a full-duplex transmission rate of 100 Mbit/s is used.
- LVDS low voltage differential signaling method
- it since it does not require a lot of processors, there is no overload, so it may not be harmful to the driver or the driver's safety.
- it is possible to reduce the size of the data transmitted by the communication network in the vehicle, so even if it is applied to an autonomous vehicle, it is possible to eliminate problems caused by the communication method and communication speed caused by the operation of a plurality of cameras arranged in the vehicle exist.
- the image sensor 1112 may transmit data after down-sampling the Bayer pattern type frame to a size of 1/n. Downsampling may be performed after smoothing through a Gaussian filter or the like on data of a Bayer pattern received before downsampling. Thereafter, after generating a frame packet based on the down-sampled image data, the completed frame packet may be transmitted to the first processing unit 1121. However, this function may be performed by the first processing unit 1121 instead of the image sensor.
- the image sensor 1112 may include a serializer (not shown) that converts Bayer patterns into serial data in order to transmit Bayer data in a serial communication method such as a low voltage differential signaling method (LVDS).
- the serializer may include or be implemented with a buffer that temporarily stores data and a phase-locked loop (PLL) that forms a period of transmitted data.
- PLL phase-locked loop
- the connector 1114 outputs data generated by the camera module 1110 to the outside or receives data from the outside.
- the connector 1114 may be formed as a communication unit, and may be formed as a communication line or a data line.
- the connector 1114 may transmit Bayer data generated and output by the image sensor 1112 to the first processing unit 1121.
- the connector 1114 formed of a line connected to the outside may be implemented as a Mobile Industry Processor Interface (MIPI).
- MIPI is an interface between each component constituting a mobile device, and includes a display serial interface (DSI) and a camera serial interface (CSI) as interfaces with enhanced reusability and compatibility.
- DSI display serial interface
- CSI camera serial interface
- the connector 1114 of the camera module 1110 may be implemented with CSI.
- the camera module 1110 may further include a driver IC and an actuator.
- Driver IC refers to a semiconductor (IC) that provides driving signals and data to a panel as electric signals to display text or video images on a screen, and may be disposed at various locations of a mobile terminal device. Also, the driver IC can drive the actuator.
- the actuator may adjust the focus by adjusting the position of the lens or the barrel including the lens.
- the actuator may be a VCM (Voice Coil Motor) type.
- the lens may include a variable focus lens.
- the driver IC may drive the varifocal lens.
- the lens may include a liquid lens containing a liquid, and in this case, the driver IC may adjust the focus by adjusting the liquid in the liquid lens.
- the AP module 1120 receives first output data output from the camera module 1110.
- the AP module 1120 receives first Bayer data output from the image sensor 1112 from the camera module 1110.
- the AP Application Processor 120
- the AP is a mobile memory chip, and when the image processing device 1130 is a mobile device, it refers to a core semiconductor that operates various applications and processes graphics in the mobile device.
- the AP module 1120 will be implemented in the form of a System on Chip (SoC) that includes all the functions of the computer's central processing unit (CPU) and the functions of a chipset that controls the connection of other equipment such as memory, hard disk, and graphics card. I can.
- SoC System on Chip
- the AP module 1120 includes a first processing unit 1121 and a second processing unit 1122.
- the first processing unit 1121 generates second Bayer data having a second resolution by using the first Bayer data having a first resolution.
- the first processing unit 1121 increases the resolution of Bayer data, which is image data generated and output by the image sensor 1112. That is, the second Bayer data of the second resolution is generated from the first Bayer data of the first resolution.
- the second resolution means a resolution having a resolution value different from the first resolution, and the second resolution may be higher than the first resolution.
- the first resolution may be the resolution of Bayer data output from the camera module 1110, and the second resolution may be changed according to a user's setting or may be a preset resolution.
- the image sensor 1112 may be an RGB image sensor.
- the image processing apparatus 1130 may further include an input unit (not shown) that receives information on resolution from a user.
- the user may input information on the second resolution to be generated by the first processing unit 1121 through the input unit.
- the second resolution can be set to a resolution that is different from the first resolution, and if a new image is to be acquired within a relatively short time, the difference from the first resolution is The second resolution can be set to a resolution that is not much.
- the first processing unit 1121 may generate second Bayer data having a second resolution from first Bayer data having a first resolution in order to perform a super resolution (SR).
- Super resolution is a process of generating a high-resolution image based on a low-resolution image. It functions as a digital zoom that generates a high-resolution image from a low-resolution image through image processing rather than a physical optical zoom. Super resolution may be used to improve the quality of a compressed or down-sampled image, or may be used to improve the quality of an image of resolution according to device limitations. In addition, it can be used to increase the resolution of images in various fields.
- Bayer data is raw data generated and output by the image sensor 1112 and includes more information than an RGB image generated by performing image processing.
- RGB data in 10-bit or more, but it is currently applied only to displays, and when using existing RGB data, each channel has 8-bit data. In addition, information may be lost due to demosaic performed at the ISP level.
- Bayer data has 10-bit data in the unprocessed format currently used in mobile. Bayer data in 12-bit or 14-bit format is also possible. As described above, in the case of using Bayer data, since the amount of input information to be used for super resolution increases compared to RGB data, increasing the resolution using Bayer data has superior processing quality compared to increasing the resolution using an RGB image.
- the first processing unit 1121 may increase the resolution of IR data as well as Bayer data.
- IR data of a fourth resolution may be generated using IR data of a third resolution generated by the ToF sensor and output from the camera module 1110.
- the third resolution may be a resolution of IR data output from the ToF sensor 1120, and the fourth resolution may be changed according to a user's setting or may be a preset resolution.
- the fourth resolution may be a resolution having the same resolution value as the second resolution.
- the fourth resolution of the IR data is equal to the second Bayer data so that the size of the IR image and the RGB image, that is, the resolution are the same.
- the branch can generate IR data to be the same as the second resolution.
- the second processing unit 1122 performs image processing by receiving second output data output from the first processing unit 1121.
- the second processing unit 1122 generates an image by performing image signal processing (ISP) on the second output data output from the first processing unit 1121.
- the second processing unit 1122 may be an image signal processor (ISP).
- the second output data output from the first processing unit 1121 may be received using MIPI (Mobile Industry Processor Interface) communication, and an image signal processing process may be performed.
- MIPI Mobile Industry Processor Interface
- the second processor 1122 may generate an RGB image from the second Bayer data.
- the second processing unit 1122 may perform a plurality of sub-processes while processing the image signal. For example, any one or more of a gamma correction, color correction, auto exposure correction, and auto white balance process may be performed on the received image. .
- the second processing unit 1122 is the RGB image generated from Bayer data, which is the second output data of the first processing unit 1121, and IR data.
- an RGB image with improved image quality can be generated.
- An RGB image generated only with Bayer data in a low-light environment has low brightness or severe noise, resulting in a lot of deterioration in image quality.
- an IR image may be used. That is, by calculating the RGB image and the IR image, the second RGB image with improved image quality may be generated.
- the camera module 1110 including the RGB image sensor and the ToF image sensor, it is possible to improve the low-light intensity of the RGB image by using high-resolution IR data as well as a zoom function that increases the resolution of each data.
- Bayer data or IR data may generate a high-resolution RGB image, a high-resolution IR image, and a high-resolution depth image through a process of increasing the resolution.
- a processing unit that processes IR data in high resolution is suitable to be implemented in a chip form.
- the process of increasing the resolution of IR data may use a chip of the first processing unit 1121 that increases the resolution of Bayer data. It is only necessary to switch the learned weight values to increase the resolution of IR data while using a part of the first processing unit 1121 chip. If the RGB image in low-light conditions is improved by using the IR image with improved resolution in this way, a higher improvement effect can be obtained, and various applications (e.g. face recognition, object recognition, size) through depth image and fusion Recognition, etc.), the recognition rate is improved.
- the first processing unit 1121 may generate the second Bayer data from the first Bayer data.
- the meaning of performing deep learning means generating the second Bayer data using a convolutional neural network derived through deep learning.
- the second Bayer data may be generated from the first Bayer data by using an algorithm that increases resolution other than deep learning. It is natural that various algorithms used in Super Resolution (SR) can be used. The process of increasing the resolution of the first output data by the first processing unit 1121 will be described in detail later with reference to FIGS. 18 and 3 to 7.
- the first processing unit 1121 may be implemented in a chip form separate from the second processing unit 1122.
- the first processing unit 1121 for generating second Bayer data from the first Bayer data and the second processing unit 1122 for performing image signal processing may be implemented as separate chips.
- the camera module 1110 may be a camera device included in the mobile device, and an AP module 1120 processing various applications of the mobile device is formed.
- the first processing unit 1121 may be implemented in the form of a separate chip separated from the second processing unit 1122, which is an ISP processor, on the AP module.
- Bayer data which is first output data generated and output by the camera module 1110, may be raw raw data, and in this case, Bayer data may be represented as Bayer raw data.
- Bayer data is received by the first processing unit 1121 formed in a chip form on the AP module through MIPI communication.
- the first processing unit 1121 generates second Bayer data from the first Bayer data using a convolutional neural network learned by performing deep learning.
- the first processing unit 1121 may be referred to as a deep learning network chip. Since the first processing unit 1121 receives and processes the low-resolution Bayer row data, it is not necessary to take into account the MIPI bandwidth or additional work on a control signal separate from the camera module. Therefore, existing devices can be used as they are, so that the compatibility is high and the degree of freedom in design is increased.
- the first processing unit 1121 generates second Bayer data using the first Bayer data, and the second output data output from the first processing unit 1121 is received by the second processing unit 1122 to perform image signal processing. To create an image.
- the first processing unit 1121 may be implemented in the form of an IP block separated from the second processing unit on the AP module.
- IP (intellectual property) block refers to a reusable logic unit, cell, or chip layout design, and refers to a block considered as intellectual property of a specific party.
- the IP block may be an IP core.
- IP blocks are used as building blocks within IC designs by parties with licensed and/or owned intellectual property (e.g., patents, source code copyrights, trade secrets, know-how, etc.) present in the design. I can.
- the IP block is a design block that can be applied to a corresponding chip for designing a chip such as an AP module, and chip design can be facilitated by using an IP block. By using multiple IP blocks, it is possible to design a chip like an SoC.
- the AP module includes IP blocks of various applications that perform functions in the mobile device.
- LTE modem Graphic Processing Unit (GPU), Wi-Fi, Display Processing Unit (DPU), Video Processing Unit (VPU), Digital Signal Processor (DSP), Hexagon Vector eXtensions (HVX), All-Ways Aware, Audio, Central A Camera Image Signal corresponding to the IP block of an application that performs functions necessary for a mobile device, such as a Processing Unit (CPU), location, and security, and a second processing unit 1122 that performs image signal processing on image signals received from the camera device. It may include a Processor (ISP).
- ISP Processor
- the first processing unit 1121 and the second processing unit 1122 may be implemented as an IP block of an AP module.
- the second processing unit 1122 may be an image signal processing (ISP) IP block.
- the first processing unit 1121 for generating second Bayer data from the first Bayer data may be configured as an IP block and added to an existing chip of the AP module.
- the IP block of the first processing unit 1121 may be formed to be distinguished from the image signal processing IP block that is the second processing unit.
- the first processing unit 1121 When the first processing unit 1121 is formed in the form of an IP block separated from the second processing unit on the AP module, Bayer data, which is the first output data output from the camera module 1110, is The first processing unit 1121 formed of an IP block receives it. Thereafter, the first processing unit 1121 IP block generates second Bayer data using the first Bayer data, and the second output data output from the first processing unit 1121 IP block is the second processing unit 1122 IP block. It receives and performs image signal processing to generate an image.
- the first processing unit 1121 includes a deep learning network (Deep Learning Network, 121-1) that generates Bayer data of a second resolution from first Bayer data of a first resolution, as shown in FIG.
- a Bayer parameter 1121-2 which is a deep learning parameter used to generate Bayer data of a second resolution, from the first Bayer data may be stored.
- the deep learning parameter 1121-2 may be stored in a memory.
- the first processing unit 1121 may be implemented in the form of a chip to generate second Bayer data from the first Bayer data.
- the first processing unit 1121 may include one or more processors, and at least one program command executed through the processor may be stored in one or more memories.
- the memory may include volatile memory such as S-RAM and D-lap.
- the present invention is not limited thereto, and in some cases, the memory 1115 may be a flash memory, a ROM (Read Only Memory), an Erasable Programmable Read Only Memory (EPROM), or an Electrically Erasable Programmable Read Only Memory (EPROM). It may also include nonvolatile memory such as.
- a typical camera device or camera module 1110 outputs image-type data through a process (color interpolation process, color interpolation or demosaic) that receives Bayer patterns from an image sensor and coats colors. ) It is possible to extract information including information and transmit data including the extracted information to the outside.
- the Bayer pattern may include raw data output by an image sensor that converts an optical signal included in the camera device or camera module 1110 into an electrical signal.
- the deep learning algorithm (model) applied to the first processing unit 1121 is an algorithm that generates image data with a resolution higher than the resolution of the input image data, and repeatedly performs learning through deep learning training. It can mean the generated optimal algorithm.
- Deep learning sometimes expressed as deep learning, is machine learning that attempts to achieve a high level of abstraction (summarizing key contents or functions in a large amount of data or complex data) through a combination of several nonlinear transformation methods. It refers to a set of algorithms for (machine learning).
- deep learning is a model for representing training data in a form that a computer can understand (for example, in the case of images, pixel information is expressed as a column vector) and applied to learning.
- a learning technique that derives can include learning techniques such as Deep Neural Networks (DNN) and Deep Belief Networks (DBN). .
- DNN Deep Neural Networks
- DBN Deep Belief Networks
- the first processing unit 1121 generates second Bayer data from the first Bayer data.
- the deep learning model of FIG. 3 may be used.
- the deep learning model of FIG. 3 is a deep learning model to which a deep neural network (DNN) algorithm is applied, and is a diagram illustrating a process of generating data of a new resolution according to the application of the DNN algorithm.
- DNN deep neural network
- a deep neural network is a deep neural network in which multiple hidden layers exist between an input layer and an output layer, and a connection pattern between neurons similar to the structure of an animal's visual cortex. It can be embodied as a convolutional neural network that forms a convolutional neural network and a recurrent neural network that builds up the neural network every moment over time.
- DNN classifies neural networks by reducing and distorting the amount of data by repeating convolution and sub-sampling.
- DNN outputs classification results through feature extraction and classification behavior, mainly used to analyze images, and convolution means image filtering.
- the first processing unit 1121 to which the DNN algorithm is applied performs deep learning.
- the first processing unit 1121 is intended to increase the magnification based on Bayer data 110 of the first resolution. Convolution and sub-sampling are performed on the region.
- Increasing the magnification means expanding only a specific portion of the first Bayer data. Accordingly, since a portion not selected by the user is a portion that the user is not interested in, there is no need to perform a process of increasing the resolution, and thus the convolution and sub-sampling process can be performed only on the portion selected by the user. Through this, by not performing unnecessary operations, it is possible to reduce the amount of calculations and thus increase the processing speed.
- Sub-sampling refers to the process of reducing the size of an image. At this time, the sub-sampling may use a Max Pool method or the like. Max-Pull is a technique that selects the maximum in a given area, similar to how neurons respond to the largest signal. Sub-sampling has the advantage of reducing noise and increasing the speed of learning.
- a plurality of image data 120 may be output as shown in FIG. 3.
- the plurality of image data 120 may be a feature map.
- a plurality of image data having different characteristics may be output using an up-scale method based on the plurality of image data.
- the up-scale method means that the image is scaled up by r*r times by using different r ⁇ 2 filters.
- the first processing unit 1121 When a plurality of image data according to the upscale is output as shown in FIG. 3 (130), the first processing unit 1121 recombines based on these image data, and finally, second Bayer data 140 of a second resolution. ) Can be printed.
- a deep learning parameter used by the first processing unit 1121 to perform deep learning to generate second Bayer data from the first Bayer data may be derived through deep learning training.
- Deep learning can be divided into training and inference.
- Training refers to a process of learning a deep learning model through input data
- Inference refers to a process of performing image processing and the like with the learned deep learning model. That is, the image is processed using a deep learning model to which parameters of a deep learning model derived through training are applied.
- a first deep learning parameter required for Bayer data processing must be derived through training.
- an inference for generating second Bayer data from the first Bayer data may be performed by performing deep learning using a deep learning model to which the corresponding Bayer parameter is applied. Therefore, a training process for deriving parameters for performing deep learning must be performed.
- the deep learning training process may be performed through repetitive learning as shown in FIG. 4. After receiving the first sample data X and the second sample data Z having different resolutions, deep learning training may be performed based on the input.
- a higher resolution based on a parameter generated by comparing and analyzing the first output data (Y) and the second sample data (Z) that performed deep learning training using the first sample data (X) as input data. You can create an algorithm that generates Bayer data.
- the first output data (Y) is data output by performing real deep learning
- the second sample data (Z) is data input by the user.
- the most It may mean data that can be output ideally.
- the first sample data X may be data obtained by down-sampling the second sample data Z to reduce the resolution.
- the degree of downsampling may vary according to a ratio to be enlarged through deep learning, that is, a zoom ratio to perform digital zoom. For example, if the zoom ratio to be performed through deep learning is 3 times and the resolution of the second sample data Z is 9 MP (Mega Pixel), the resolution of the first sampling data X must be 1 MP, and deep learning is performed. As a result, the resolution becomes 9MP of the first output data (Y) whose resolution is increased by 3 times, and the second sample data (Z) of 9M is down-sampled to 1/9 to generate 1MP of first sample data (Y). I can.
- the difference between the two data is calculated by comparing and analyzing the first output data Y and the second sample data Z output through deep learning execution according to the input of the first sample data X, and the difference between the two data. It is possible to give feedback to the parameters of the deep learning model in the direction of reducing.
- the difference between the two data may be calculated through a mean squared error (MSE) method, which is one of the loss functions.
- MSE mean squared error
- various loss functions such as CEE (Cross Entropy Error) can be used.
- the second sample data (Z) and the first output data (Y), which are actual output data are provided by giving feedback by changing or deleting the parameter or creating a new parameter. Can make no difference.
- the algorithm to which deep learning is applied through this method may derive parameters such that the first output data Y is output similarly to the second sample data Z.
- the resolution of the second sample data (Z) may be the same as or higher than the resolution of the first output data (Y), and the resolution of the second sample data (Z) may be the same as the resolution of the first output data (Y). I can.
- the output result and the comparison object exist, and training may be performed using a compensation value as well as a case in which learning is performed through comparison with the comparison object.
- the surrounding environment may be first recognized and the current environment state may be transmitted to a processor that performs deep learning training.
- the processor performs the corresponding action, and the environment again informs the processor of the reward for the action. And the processor chooses the action that maximizes the reward.
- Training can be performed by repeatedly performing learning through this process.
- deep learning training can be performed using various deep learning training methods.
- the deep learning process and the number of memory gates must be minimized.
- the factors that most affect the number of gates are algorithm complexity and clock ( It is the amount of data processed per clock, and the amount of data processed by the processor depends on the input resolution.
- the processor 1220 generates an image with a high magnification by reducing the input resolution in order to reduce the number of gates and then upscaling it later, so there is an advantage of generating an image faster. do.
- zoom 2x by upscaling the width and height by 2x each based on a 1/4 area (2Mp).
- 2Mp an input resolution
- the horizontal and vertical are respectively upscaled by 4 times based on the generated image. If you zoom 4x with the (Up scailing) method, you can create a zoomed image of the same area as the 2x zoom.
- the first processing unit 1121 since the first processing unit 1121 according to the second embodiment of the present invention applies an algorithm that has already been generated through deep learning, it can be easily applied to a low-end camera module and various devices including the same. In application to this, since high resolution is implemented by using only a few line buffers, there is also an effect of implementing a processor with a relatively small chip.
- the first processing unit 1121 includes at least one line buffer that stores the first bait data for each line, and when a predetermined number of first Bayer data is stored in the line buffer, the first bayer stored in the line buffer Second Bayer data may be generated for data.
- the first processing unit 1121 divides and receives the first Bayer data for each line, and stores the first Bayer data received for each line in a line buffer. After receiving the first Bayer data of all lines, the first processing unit 1121 does not generate the second Bayer data, and when the first Bayer data of a certain number of lines is stored, the first Bayer data stored in the line buffer is stored.
- the second Bayer data may be generated.
- the first processing unit 121 generates a plurality of line buffers 11 for receiving first Bayer data, and first array data for arranging first Bayer data output through the line buffer for each wavelength band.
- the first data alignment unit 221 that performs deep learning, the deep learning processor 222 that performs deep learning, the second array data output through the deep learning processor 222 is arranged in a Bayer pattern to generate second Bayer data. 2 It may include a plurality of line buffers 12 for outputting second Bayer data output through the data alignment unit 223 and the second data alignment unit 223.
- the first Bayer data is information including the Bayer pattern described above, and is described as Bayer data in FIG. 5, but may be defined as a Bayer image or a Bayer pattern.
- first data alignment unit 221 and the second data alignment unit 223 are illustrated as separate components for convenience, but are not limited thereto, and the deep learning processor 222 will first align the data to be described later.
- a function performed by the unit 221 and the second data alignment unit 223 may be performed.
- the first Bayer data having a first resolution may transmit image information on a region selected by a user to (n+1) line buffers 11a, 11b, ⁇ 11n. 11n+1). .
- image information for the region not selected by the user is not transmitted to the line buffer 11.
- the first Bayer data includes a plurality of row data
- the plurality of row data may be transmitted to the first data alignment unit 221 through the plurality of line buffers 11.
- the area in which deep learning is to be performed by the deep learning processor 222 is a 3 X 3 area
- a total of 3 lines must be simultaneously transmitted to the first data alignment unit 221 or the deep learning processor 222. You can run. Accordingly, information on the first line of the three lines is transmitted to the first line buffer 11a and then stored in the first line buffer 11a, and information on the second line of the three lines is transmitted to the second line buffer. After being transmitted to (11b), the second line buffer (11b) may be stored.
- the third line since there is no information on a line to be received thereafter, it is not stored in the line buffer 11 and may be immediately transmitted to the deep learning processor 222 or the first data alignment unit 221.
- the first line buffer 11a and the second line buffer 11b Information on the line and information on the second line may be simultaneously transmitted to the deep learning processor 222 or the first image alignment unit 219.
- the area in which deep learning is to be performed by the deep learning processor 222 is a (N+1) x (N+1) area
- a total of (N+1) lines are the first data alignment unit 221 or Deep learning can be performed only when it is simultaneously transmitted to the deep learning processor 222. Accordingly, information on the first line among (N+1) lines is transmitted to the first line buffer 11a, and then stored in the first line buffer 11a, and the second line among (N+1) lines
- Information on the second line buffer 11b may be stored after being transmitted to the second line buffer 11b, and information on the N-th line among (N+1) lines is the N-th line buffer 11n After being transmitted to, the Nth line buffer 11n may be stored.
- the first data alignment unit 221 or the deep learning processor 222 since the first data alignment unit 221 or the deep learning processor 222 must simultaneously receive information on N+1 lines, the first data stored in the line buffers 11a to 11n. Information on the nth line to the nth line may also be simultaneously transmitted to the deep learning processor 222 or the first image alignment unit 219.
- the first image alignment unit 219 After receiving Bayer data from the line buffer 11, the first image alignment unit 219 generates first alignment data by arranging Bayer data by wavelength band, and then converts the generated first alignment data into a deep learning processor 222 ) Can be sent.
- the first image alignment unit 219 may generate first array data arranged by classifying the received information into specific wavelengths or specific colors (Red, Green, Blue).
- the deep learning processor 222 may generate second array data by performing deep learning based on the first array data received through the first image alignment unit 219.
- the deep learning processor 222 performs deep learning based on the first array data received through the first image alignment unit 219 to generate second array data of a second resolution higher than the first resolution. can do.
- deep learning is performed for the 3 x 3 area, and the first array for the (n+1) x (n+1) area
- deep learning can be performed on the (n+1) x (n+1) region.
- the second alignment data generated by the deep learning processor 222 is transmitted to the second data alignment unit 223, and the second data alignment unit 223 converts the second alignment data to a second alignment data having a Bayer pattern. Can be converted to Bayer data.
- the converted second Bayer data is output to the outside through a plurality of line buffers 12a, and the output second Bayer data is Bayer data having a second resolution that is higher than the first resolution by another process.
- FIGS. 6 and 7 are diagrams illustrating an image in which an image of first Bayer data having a first resolution is converted into second Bayer data having a second resolution by the first processing unit 1121.
- the first processing unit 1121 converts the resolution for the region, and as a result, as shown in FIGS. 6 and 7, Bayer data 40 may be generated.
- the first processing unit 1121 may pre-process the second Bayer data.
- the first processing unit 1121 may generate second Bayer data having a second resolution by using the first Bayer data having a first resolution, and perform preprocessing on the generated second Bayer data.
- the first processing unit 1121 is at least one of white balance, de-nosing, de-focus, de-blur, and di-mosaic. You can do the above.
- various pretreatments corresponding to pretreatment may be performed.
- the amount of computation of the second processing unit 1122 can be reduced by performing not only a super resolution that increases the resolution of Bayer data in the first processing unit 1121 but also pre-processing an image that can be processed by the second processing unit 1122. That is, the first processing unit 1121 performs the pre-processing function of the second processing unit 1122 and shares the functions, thereby reducing the burden on the ISP.
- the first processing unit 1121 may further include a pre-processing unit that performs pre-processing.
- the first processing unit 1121 may pre-process the second Bayer data using a convolutional neural network learned by performing deep learning. If there is a preprocessing process that can be processed using the same deep learning network as the process of increasing the resolution of Bayer data among the preprocessing process performed by the first processing unit 1121, the deep learning parameters for the preprocessing process are stored, and the corresponding Preprocessing can be performed using deep learning parameters. It may also include a separate deep learning network for one or more pre-processing processes and a memory for storing corresponding deep learning parameters.
- the first processing unit 1121 may generate an RGB image or a ycbcr image by preprocessing the second bait data.
- the preprocessed third Bayer data may be generated by performing preprocessing on the second Bayer data having the second resolution, or further, an RGB image or a ycbcr image converted from the RGB image may be generated through preprocessing.
- the first processing unit 1121 performs a plurality of pre-processing processes, but may output various data according to the performed pre-processing process. That is, it is possible to generate third Bayer data before the RGB image from the second Bayer data.
- the third Bayer data is not an RGB image, but third Bayer data, which is Bayer data on which white balance has been performed, may be generated through preprocessing such as white balance.
- an RGB image may be generated by performing preprocessing for generating an RGB image on the second Bayer data.
- a ycbcr image may be generated through ycbcr conversion of the RGB image generated as described above.
- the ycbcr image can be displayed directly on the display.
- the pre-processing process performed by the first processing unit 1121 may vary according to a user's setting, a use environment, or a working state of the ISP, which is the second processing unit 1122.
- the first processing unit 1121 performs pre-processing and shares the function of the ISP, which is the second processing unit 1122, and which pre-processing process is performed by the first processing unit 1121 may be set according to a user's setting.
- the first processing unit 1121 may set which preprocessing process to be performed. This may receive environment information from one or more processors to determine a preprocessing setting value in the first processing unit 1121.
- setting values according to environment information, etc. may be stored in a lookup table LUT, and the setting values for performing preprocessing in the first processing unit 1121 may be applied.
- the image processing apparatus 2100 may include a camera module 2110, a Bayer data processing module 2120, and an AP module 2130, as shown in FIG. 19.
- the first processing unit 1121 of the image processing apparatus 1130 of FIG. 13 or 16 may be configured as a Bayer data processing module 2120 that is a separate module other than the AP module. Except for implementing the Bayer data processing module 2120 as a separate module from the AP module 2130, the image processing process performed by the image processing apparatus 2100 of FIG. 19 is performed by the image processing apparatus 1130 of FIG.
- descriptions of the image processing apparatus 2100 of FIG. 19 that overlap with the image processing process performed by the image processing apparatus 1130 of FIG. 13 will be omitted and briefly described below.
- the camera module 2110 includes an image sensor, and the Bayer data processing module 2120 generates second Bayer data of a second resolution using first Bayer data of a first resolution output from the camera module 2110. .
- the Bayer data processing module 2120 may be implemented in the form of a separate chip separated from the AP module 2130.
- the Bayer data processing module 2120 may generate the second Bayer data from the first Bayer data, and the Bayer data processing module 2120 may include a preprocessor for preprocessing the second Bayer data.
- the preprocessor may generate any one of third Bayer data, RGB image, or ycbcr image by preprocessing the second Bayer data.
- the AP module 2130 performs image processing by receiving output data output from the Bayer data processing module 2120.
- An image processing apparatus includes a first processing unit that generates second Bayer data of a second resolution by using first Bayer data of a first resolution, wherein the first processing unit is It may be formed to be separated from the formed image signal processing unit.
- the first processing unit 1121 is a configuration corresponding to the first processing unit 1121 of the image processing apparatus 1130 of FIG. 13 or 16, and the Bayer data processing module 2120 of the image processing apparatus 2100 of FIG. 19
- the image processing apparatus according to another embodiment of the present invention may include a first processing unit, and the first processing unit may generate the second Bayer data from the first Bayer data, and the second Bayer data It may include a pre-treatment unit for pre-processing.
- the preprocessor may generate any one of third Bayer data, RGB image, or ycbcr image by preprocessing the second Bayer data.
- FIG. 20 is a flowchart of an image processing method according to a second embodiment of the present invention
- FIG. 21 is a flowchart of an image processing method according to another embodiment of the present invention.
- a detailed description of each step of FIGS. 20 to 21 corresponds to a detailed description of the image processing apparatus of FIGS. 13 to 19, and redundant descriptions will be omitted below.
- An image processing method relates to a method of processing an image in an image processing apparatus including one or more processors.
- step S11 first Bayer data having a first resolution is received, and second Bayer data having a second resolution is generated in step S13 using a convolutional neural network learned by performing deep learning in step S12.
- step S13 it is possible to perform pre-processing on the second Bayer data in step S21, and after performing the pre-processing, the step of generating any one of the third Bayer data, RGB image, or ycbcr image in step S22. I can. Thereafter, the step of generating an image that can be output to the display through image signal processing may be further included.
- Modifications according to the present embodiment may include some configurations of the first embodiment described with reference to FIGS. 1 to 12 and some configurations of the second embodiment described with reference to FIGS. 13 to 21. That is, modified examples include the embodiments described with reference to FIGS. 1 to 12, but some configurations of the embodiments described with reference to FIGS. 1 to 12 are omitted, and some configurations of the embodiments described with reference to corresponding FIGS. 13 to 21 It may include. Alternatively, the modified example may include some configurations of the embodiments described with reference to the corresponding FIGS. 1 to 12, and some configurations of the embodiments described with reference to FIGS. 13 to 21 are omitted.
- the embodiments of the present invention can be implemented as computer-readable codes on a computer-readable recording medium.
- the computer-readable recording medium includes all types of recording devices that store data that can be read by a computer system.
- Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tapes, floppy disks, optical data storage devices, and the like.
- computer-readable recording media are distributed across networked computer systems. In a distributed manner, computer-readable code can be stored and executed.
- functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the technical field to which the present invention belongs.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Studio Devices (AREA)
- Image Processing (AREA)
- Color Television Image Signal Generators (AREA)
Abstract
Description
Claims (10)
- 제1 해상도를 가지는 제1 베이어 데이터를 생성하는 이미지 센서; 및상기 제1 베이어 데이터를 이용하여 제2 해상도를 가지는 제2 베이어 데이터를 출력하는 프로세서를 포함하는 카메라 장치.
- 제1항에 있어서,상기 프로세서는,제1 베이어 데이터를 이용하여 제2 해상도를 가지는 제2 베이어 데이터를 출력하도록 학습된 컨볼루션 신경망을 포함하는 카메라 장치.
- 제2항에 있어서,상기 컨볼루션 신경망의 트레이닝 세트는,제1 해상도를 가지는 제1 베이어 데이터 및 제2 해상도를 가지는 제2 베이어 데이터를 포함하는 카메라 장치.
- 제1항에 있어서,상기 제2 해상도는 상기 제1 해상도보다 높은 카메라 장치.
- 제1항에 있어서,상기 제2 베이어 데이터를 Image Signal Processor로 출력하는 카메라 장치.
- 제1항에 있어서,상기 프로세서는,상기 제1 베이어 데이터를 수신하는 수신부; 및상기 제1 베이어 데이터를 이용하여 제2 해상도를 가지는 제2 베이어 데이터를 출력하는 콘볼루션 신경망을 포함하는 카메라 장치.
- 제1 해상도를 가지는 제1 베이어 데이터를 입력받는 단계; 및학습된 콘볼루션 신경망을 이용하여 상기 제1 베이어 데이터로부터 제2 해상도를 가지는 제2 베이어 데이터를 출력하는 단계를 포함하는 방법.
- 제7항에 있어서,상기 제1 베이어 데이터는 이미지 센서로부터 출력되는 데이터인 방법.
- 제7항에 있어서,상기 컨볼루션 신경망의 트레이닝 세트는,제1 해상도를 가지는 제1 베이어 데이터 및 제2 해상도를 가지는 제2 베이어 데이터를 포함하는 방법.
- 제7항에 있어서,상기 제2 베이어 데이터를 Image Signal Processor로 출력하는 방법.
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2021572932A JP2022536327A (ja) | 2019-06-13 | 2020-06-15 | カメラ装置及びカメラ装置のイメージ生成方法 |
EP20823426.0A EP3985961A4 (en) | 2019-06-13 | 2020-06-15 | CAMERA DEVICE AND METHOD FOR GENERATING CAMERA DEVICE IMAGES |
US17/617,832 US20220253978A1 (en) | 2019-06-13 | 2020-06-15 | Camera device and image generation method of camera device |
CN202080057178.3A CN114270799B (zh) | 2019-06-13 | 2020-06-15 | 相机装置以及相机装置的图像生成方法 |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020190070298A KR102242939B1 (ko) | 2019-06-13 | 2019-06-13 | 카메라 장치 및 카메라 장치의 이미지 생성 방법 |
KR10-2019-0070298 | 2019-06-13 | ||
KR10-2019-0130677 | 2019-10-21 | ||
KR1020190130677A KR20210047070A (ko) | 2019-10-21 | 2019-10-21 | 이미지 처리 장치 및 이미지 처리 방법 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2020251337A1 true WO2020251337A1 (ko) | 2020-12-17 |
Family
ID=73781462
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2020/007738 WO2020251337A1 (ko) | 2019-06-13 | 2020-06-15 | 카메라 장치 및 카메라 장치의 이미지 생성 방법 |
Country Status (6)
Country | Link |
---|---|
US (1) | US20220253978A1 (ko) |
EP (1) | EP3985961A4 (ko) |
JP (1) | JP2022536327A (ko) |
CN (1) | CN114270799B (ko) |
TW (1) | TW202105028A (ko) |
WO (1) | WO2020251337A1 (ko) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20030020357A (ko) * | 2000-07-06 | 2003-03-08 | 더 트러스티스 오브 콜롬비아 유니버시티 인 더 시티 오브 뉴욕 | 데이터 해상도를 향상시키는 방법 및 장치 |
KR101791573B1 (ko) * | 2016-10-21 | 2017-10-31 | 서강대학교산학협력단 | 컨볼루션 신경망을 이용한 비디오 스트림에 대한 수퍼 해상도 장치 및 방법 |
JP2018151747A (ja) * | 2017-03-10 | 2018-09-27 | 株式会社ツバサファクトリー | 超解像度処理装置、超解像度処理方法およびコンピュータプログラム |
KR20180122548A (ko) * | 2017-05-03 | 2018-11-13 | 삼성전자주식회사 | 이미지를 처리하는 방법 및 장치 |
KR20190021756A (ko) * | 2017-08-23 | 2019-03-06 | 엘지디스플레이 주식회사 | 영상 처리 방법 및 이를 이용한 표시 장치 |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20180008244A (ko) * | 2016-07-14 | 2018-01-24 | 엘지이노텍 주식회사 | 이미지 생성 방법 및 생성 장치 |
US10643306B2 (en) * | 2017-10-11 | 2020-05-05 | Qualcomm Incoporated | Image signal processor for processing images |
US10733714B2 (en) * | 2017-11-09 | 2020-08-04 | Samsung Electronics Co., Ltd | Method and apparatus for video super resolution using convolutional neural network with two-stage motion compensation |
CN109658348A (zh) * | 2018-11-16 | 2019-04-19 | 天津大学 | 基于深度学习的联合噪声估计和图像去噪方法 |
-
2020
- 2020-06-15 WO PCT/KR2020/007738 patent/WO2020251337A1/ko active Application Filing
- 2020-06-15 EP EP20823426.0A patent/EP3985961A4/en active Pending
- 2020-06-15 TW TW109120316A patent/TW202105028A/zh unknown
- 2020-06-15 JP JP2021572932A patent/JP2022536327A/ja active Pending
- 2020-06-15 US US17/617,832 patent/US20220253978A1/en not_active Abandoned
- 2020-06-15 CN CN202080057178.3A patent/CN114270799B/zh active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20030020357A (ko) * | 2000-07-06 | 2003-03-08 | 더 트러스티스 오브 콜롬비아 유니버시티 인 더 시티 오브 뉴욕 | 데이터 해상도를 향상시키는 방법 및 장치 |
KR101791573B1 (ko) * | 2016-10-21 | 2017-10-31 | 서강대학교산학협력단 | 컨볼루션 신경망을 이용한 비디오 스트림에 대한 수퍼 해상도 장치 및 방법 |
JP2018151747A (ja) * | 2017-03-10 | 2018-09-27 | 株式会社ツバサファクトリー | 超解像度処理装置、超解像度処理方法およびコンピュータプログラム |
KR20180122548A (ko) * | 2017-05-03 | 2018-11-13 | 삼성전자주식회사 | 이미지를 처리하는 방법 및 장치 |
KR20190021756A (ko) * | 2017-08-23 | 2019-03-06 | 엘지디스플레이 주식회사 | 영상 처리 방법 및 이를 이용한 표시 장치 |
Non-Patent Citations (1)
Title |
---|
See also references of EP3985961A4 * |
Also Published As
Publication number | Publication date |
---|---|
CN114270799A (zh) | 2022-04-01 |
US20220253978A1 (en) | 2022-08-11 |
CN114270799B (zh) | 2024-04-05 |
JP2022536327A (ja) | 2022-08-15 |
TW202105028A (zh) | 2021-02-01 |
EP3985961A1 (en) | 2022-04-20 |
EP3985961A4 (en) | 2023-05-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021230708A1 (en) | Image processing method, electronic device and readable storage medium | |
WO2016013902A1 (en) | Image photographing apparatus and image photographing method | |
WO2020050686A1 (en) | Image processing device and method | |
WO2017018612A1 (en) | Method and electronic device for stabilizing video | |
WO2021006366A1 (ko) | 디스플레이 패널의 색상을 조정하는 인공 지능 장치 및 그 방법 | |
WO2013147488A1 (en) | Image processing apparatus and method of camera device | |
WO2015141925A1 (en) | Photographing apparatus, method of controlling the same, and computer-readable recording medium | |
WO2021107406A1 (en) | Electronic device, control method thereof, and system | |
WO2019017698A1 (ko) | 전자 장치 및 전자 장치에서 하이 다이나믹 레인지 이미지 데이터를 압축하는 방법 | |
WO2019017641A1 (ko) | 전자 장치 및 전자 장치의 이미지 압축 방법 | |
WO2022102972A1 (ko) | 이미지 센서를 포함하는 전자 장치 및 그 동작 방법 | |
WO2021118111A1 (ko) | 이미지 처리 장치 및 이미지 처리 방법 | |
WO2022010122A1 (ko) | 영상을 제공하는 방법 및 이를 지원하는 전자 장치 | |
WO2022039424A1 (ko) | 흔들림을 보정하는 방법 및 그 전자 장치 | |
WO2022154387A1 (ko) | 전자 장치 및 그의 동작 방법 | |
WO2022146100A1 (en) | Image sharpening | |
WO2016017906A1 (ko) | 디스플레이 장치, 디스플레이 보정 장치, 디스플레이 보정 시스템 및 디스플레이 보정 방법 | |
WO2020017936A1 (ko) | 전자 장치 및 이미지의 전송 상태에 기반하여 이미지를 보정하는 방법 | |
WO2022071647A1 (en) | Video quality assessment method and apparatus | |
WO2017014404A1 (en) | Digital photographing apparatus and digital photographing method | |
WO2020251337A1 (ko) | 카메라 장치 및 카메라 장치의 이미지 생성 방법 | |
WO2020145744A1 (ko) | 카메라 장치 및 이를 구비하는 전자 장치 | |
WO2020251336A1 (ko) | 카메라 장치 및 카메라 장치의 이미지 생성 방법 | |
WO2021075799A1 (ko) | 이미지 처리 장치 및 이미지 처리 방법 | |
WO2022005002A1 (ko) | 이미지 센서를 포함하는 전자 장치 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20823426 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2021572932 Country of ref document: JP Kind code of ref document: A |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2020823426 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2020823426 Country of ref document: EP Effective date: 20220113 |