CN106296616A - A kind of infrared image detail enhancing method and a kind of infrared image details intensifier - Google Patents
A kind of infrared image detail enhancing method and a kind of infrared image details intensifier Download PDFInfo
- Publication number
- CN106296616A CN106296616A CN201610684865.3A CN201610684865A CN106296616A CN 106296616 A CN106296616 A CN 106296616A CN 201610684865 A CN201610684865 A CN 201610684865A CN 106296616 A CN106296616 A CN 106296616A
- Authority
- CN
- China
- Prior art keywords
- image
- gpu
- view data
- arm
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 156
- 230000002708 enhancing effect Effects 0.000 title claims abstract description 12
- 230000008569 process Effects 0.000 claims abstract description 115
- 238000012545 processing Methods 0.000 claims abstract description 71
- 230000004927 fusion Effects 0.000 claims abstract description 15
- 238000001914 filtration Methods 0.000 claims description 19
- 238000009826 distribution Methods 0.000 claims description 12
- 238000002156 mixing Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 8
- 230000015654 memory Effects 0.000 description 16
- 238000013507 mapping Methods 0.000 description 9
- 238000012546 transfer Methods 0.000 description 5
- 230000006835 compression Effects 0.000 description 4
- 238000007906 compression Methods 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000003139 buffering effect Effects 0.000 description 2
- 238000000151 deposition Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 150000003839 salts Chemical class 0.000 description 2
- 238000001931 thermography Methods 0.000 description 2
- 235000008534 Capsicum annuum var annuum Nutrition 0.000 description 1
- 240000008384 Capsicum annuum var. annuum Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 239000006002 Pepper Substances 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013506 data mapping Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000422 nocturnal effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000008844 regulatory mechanism Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of infrared image detail enhancing method and a kind of infrared image details intensifier, utilize the ARM in SOC framework to realize the view data collected and be filtered successively processing and gray proces, obtain the first process image；Process image to first carry out contrast process successively and carry out rectangular histogram interpolation processing, obtain the second process image；Utilize the GPU in SOC framework to process image to second and be sharpened process, obtain the 3rd process image；Processing image by the 3rd and the first process image carries out fusion treatment, the view data after process is i.e. the infrared image after obtaining image detail enhancement process.At utmost retaining the details of infrared image, picture contrast is high, and local detail is clear；And allow equipment operator in the most significant scene of variations in temperature, it is also possible to see infrared image details clearly, and, utilize SOC framework can promote the efficiency of image procossing.
Description
Technical field
The present invention relates to a kind of infrared image detail enhancing method and a kind of infrared image details intensifier, belong to infrared
Image processing field.
Background technology
The one of infrared image acquiring equipment is big, and feature is to have the highest dynamic range (such as ground and sky), and some mesh
Mark the most relatively small with the temperature difference of its background or target local.Therefore to it is red to quantify Larger Dynamic with sufficiently high precision
Outer scene, high-performance thermal imaging system the most all uses the AD of 14bits or higher precision to sample detector output signal
And quantization.And common display device or quick process only require 8bits data width, so the high accuracy data of 14bits needs
To process through overcompression.
At present, the compression method such as conventional Linear Mapping (such as AGC) or nonlinear mapping (such as histogram equalization), the most not
Can solve the Larger Dynamic image information display problem detected, thus Larger Dynamic compression of images is likely to result in original letter
Breath is lost, and shows as the loss of image detail in display image so that these methods are universal in the compression of Larger Dynamic image
There is this defect.It is a kind of advanced nonlinear image enhancement processing algorithm that image detail strengthens (DDE) technology, can retain height
Details in dynamic image, solves to position a difficult problem for low contrast target in HDR scene.
Compared to traditional linearity and nonlinear mapping compression method, DDE can at utmost retain infrared image
Details, picture contrast is high, and local detail is clear.Based on above-mentioned Performance Characteristics, DDE technology allows equipment operator in variations in temperature
In the most significant scene, it is also possible to see infrared image details clearly.The 14bits view data of Infrared Detectors collection is being entered
When row DDE processes, generally utilize FPGA or DSP architecture to realize, but the treatment effeciency of this implementation is relatively low.
Summary of the invention
It is an object of the invention to provide a kind of infrared image detail enhancing method and a kind of infrared image details intensifier,
In order to solve in the relatively low problem of existing DDE method treatment effeciency.
For achieving the above object, the solution of the present invention includes a kind of infrared image detail enhancing method, utilizes SOC framework pair
Infrared image processes, and this infrared image detail enhancing method is:
The ARM in SOC framework is utilized to carry out following operation:
It is filtered processing and carrying out at gray scale according to gray scale stretching algorithm according to filtering budget law successively to view data
Reason, obtains the first process image；
Process image to described first and carry out contrast process and according to local according to restriction contrast equalization algorithm successively
Rectangular histogram bilinear interpolation algorithm carries out rectangular histogram interpolation processing, obtains the second process image；
The GPU in SOC framework is utilized to carry out following operation:
Process image to described second and be sharpened process according to high frequency sharpening algorithm, obtain the 3rd process image；By institute
State the 3rd process image and the first process image carries out fusion treatment according to local histogram's blending algorithm, the picture number after process
According to being i.e. the infrared image after obtaining image detail enhancement process.
Data transfer mode between ARM and GPU is: ARM sends data to the buffer queue pre-build, GPU from
The described buffer queue pre-build asynchronous acquisition data.
The means realizing according to high frequency sharpening algorithm, described second process image is sharpened process are: process second
Image is divided into several image blocks, and is the described second thread bundle processing that image distribution is identical with described image block number, figure
As block and thread bundle one_to_one corresponding, with thread bundle for processing unit, the described second each image block processing image is sharpened
Process.
A kind of infrared image details intensifier, including ARM and GPU,
Described ARM performs following processing module:
Filtering Processing module, for being filtered processing according to filtering budget law to view data；
Gradation processing module, for carrying out gray scale to the view data after described Filtering Processing according to gray scale stretching algorithm
Process；
Contrast processing module, is used for the view data after described gray proces according to limiting contrast equalization algorithm
Carry out contrast process；
Rectangular histogram interpolation processing module, the view data after processing degree by contrast is according to local histogram's bilinearity
Interpolation algorithm carries out rectangular histogram interpolation processing；
Described GPU performs following processing module:
Edge contrast module, for entering according to high frequency sharpening algorithm the view data after described rectangular histogram interpolation processing
Row Edge contrast；
Rectangular histogram Fusion Module, for by the view data after described gray proces and the figure after described Edge contrast
As data carry out carrying out fusion treatment according to local histogram's blending algorithm, the view data after process is i.e. to obtain image detail
Infrared image after enhancement process.
Data transfer mode between ARM and GPU is: ARM sends data to the buffer queue pre-build, GPU from
The described buffer queue pre-build asynchronous acquisition data.
Described Edge contrast module realizes being sharpened the view data after described gray proces the means of process:
View data after described gray proces is divided into several image blocks, for described view data after described gray proces
Distribute the thread bundle identical with described image block number, image block and thread bundle one_to_one corresponding, with thread bundle for processing unit pair
Each image block of described view data after described gray proces is sharpened process.
First, in the infrared image detail enhancing method that the present invention provides, by view data being filtered process, ash
After degree process, contrast process, rectangular histogram interpolation processing, Edge contrast and rectangular histogram fusion treatment, obtain image detail and increase
Infrared image after the reason of strength.The method can at utmost retain the details of infrared image, and picture contrast is high, local detail
Clearly；And allow equipment operator in the most significant scene of variations in temperature, it is also possible to see infrared image details clearly.So,
This Enhancement Method can retain the details in high dynamic range images, solves to position low contrast mesh in HDR scene
A target difficult problem.
Further, the infrared image details enhancement method of the present invention utilizes SOC framework to realize, and utilizes in SOC framework
ARM and GPU, both cooperate with each other this image enchancing method of common implementing, and ARM carries out the filter preprocessing of next frame, gray scale
Stretching, limit contrast equilibrium, local histogram bilinear interpolation processes, high frequency is sharpened and local histogram merge part by
GPU performs, thus the idle GPU resource made full use of, by can be parallel part use suitable mode to be performed by GPU, fill
Divide and utilize the advantage that GPU computing unit is many, degree of parallelism is high, evade GPU and be bad at the weak of the process computing that branch is many, dependency is high
Point, makes full use of calculating resource, it is achieved thereby that the load balancing of ARM, GPU.The design of this Heterogeneous Computing frame by frame alleviates ARM
Load, improve execution efficiency, on the other hand, increase system overall throughput, topmost calculating is put into GPU on reality
Existing, improve performance more easily.So, this image enhaucament mode can greatly promote the efficiency of image procossing.
Accompanying drawing explanation
Fig. 1 is the overall flow figure of this image enchancing method；
Fig. 2 is the connection diagram of intensifier based on ARM and GPU and external equipment；
Fig. 3 is the schematic flow sheet of the distribution of ARM thread and filter preprocessing；
Fig. 4 is the schematic flow sheet of the distribution of ARM thread and variable-gain linear stretch；
Fig. 5 is the schematic flow sheet of the distribution of ARM thread and limit value local histogram equalization；
Fig. 6 is to use GPU to optimize the thread mapping relations schematic diagram that high frequency sharpens；
Fig. 7 is the schematic flow sheet of the distribution of ARM thread and local histogram's bilinear interpolation；
Fig. 8 is that the distribution of GPU thread sharpens and the schematic flow sheet of additive fusion with high frequency；
Fig. 9 is the overlap mode schematic diagram that GPU uploaded, downloaded, performs the time.
Detailed description of the invention
The method that the present invention provides, based on being configured with the embedded SOC framework electronic equipment of ARM and GPU, i.e. utilizes SOC frame
Structure realizes the DDE method that the present invention provides.In actual applications, it is real that the method that the present invention provides can be used for high-performance thermal imaging
Time gather transmission display field, such as monitoring, remote sensing, mapping, military surveillance etc.；The most significant in nocturnal temperature change especially
In scene, it is possible at utmost retain image detail, be effectively improved picture contrast.
As it is shown in figure 1, a kind of based on ARM and GPU the infrared DDE method that the present invention provides, it is applied to be configured with central authorities
Processor ARM and the embedded SOC framework electronic equipment of graphic process unit GPU, as it is shown in figure 1, the method comprises the following steps:
First, receiving view data, wherein view data is the 14bits view data gathered through Infrared Detectors；
Then, the central processing unit ARM in SOC framework is utilized to carry out following operation:
It is filtered processing and entering according to gray scale stretching algorithm according to filtering budget law successively to the view data received
Row gray proces, obtains the first result；
Process image to first and carry out contrast process and according to local Nogata according to restriction contrast equalization algorithm successively
Figure bilinear interpolation algorithm carries out rectangular histogram interpolation processing, obtains the second result；
Graphic process unit GPU in SOC framework is utilized to carry out following operation:
Second result is sharpened process according to high frequency sharpening algorithm, obtains the 3rd result；At the 3rd
Reason result and the first result carry out fusion treatment according to local histogram's blending algorithm, and the view data after process is i.e.
Infrared image after image detail enhancement process.
So, owing to above-mentioned image processing process is soft according to arrange in central processing unit ARM or graphic process unit GPU
Part program is carried out, so this infrared DDE method can be described as: view data order is entered by the ARM calling electronic equipment
Row filtering budget law processes, gray scale stretching algorithm process, obtains the first result, then order carries out limiting contrast and all accounts
Method and local histogram's bilinear interpolation algorithm process, obtain the second result；The GPU calling electronic equipment processes second
Result carries out high frequency sharpening algorithm process, obtains the 3rd result, and the 3rd result and the first result are carried out office
Portion's rectangular histogram blending algorithm processes, and obtains the infrared image after DDE processes.
For the ease of intuitivism apprehension, in conjunction with Fig. 1, the execution flow process of the inventive method is illustrated, Fig. 1 and below
The CPU related in accompanying drawing is all processor ARM, as it is shown in figure 1, ARM carries out mean filter, intermediate value filter to view data order
Ripple pretreatment and 14 to 10 the linear stretch processings of gray scale, obtain the first result；Again the first result order is entered
Row limits contrast equalization algorithm and local histogram's bilinear interpolation algorithm process.
It is the schematic flow sheet of each step of this image enchancing method as illustrated in figs. 2 through 8, below in conjunction with the accompanying drawings to the party
Each step in method is described in detail.
In the present embodiment, being directed to be configured with the embedded SOC framework electronic equipment of multinuclear ARM, the present invention can be same
Time multiple threads view data, and be each road view data distribution ARM kernel according to data parallel degree and system hardware performance,
So that the hardware resources such as ARM are fully used, it is to avoid undue idle situation occurs, improves overall data throughput；This
Outward, the present invention can build multichannel image parallel ARM and GPU isomery framework frame by frame, real time processed images data, more meets reality
The scene of border application, as shown in Figure 2.
As it is shown on figure 3, ARM obtains filtering in advance after view data order is carried out mean filter and medium filtering pretreatment
Process view data.Wherein, what " spiced salt " described is the form of noise, and " green pepper " is stain, and " salt " is white point, " salt-pepper noise " just
It it is the pixel noise that black white occurs on image at random.
And then, as shown in Figure 4, filter preprocessing view data is carried out at 14 to 10 gray scale linear stretches by ARM
Reason, obtains the first result.
The second process is acquired afterwards as it can be seen in figures 5 and 6, correspondingly process according to the first result for ARM
The flow chart of result, particularly as follows: to first process image successively according to limit contrast equalization algorithm carry out contrast process and
Carry out rectangular histogram interpolation processing according to local histogram's bilinear interpolation algorithm, obtain the second result.
So, can finally be obtained by ARM and carry out at rectangular histogram interpolation according to local histogram's bilinear interpolation algorithm
The result managed and obtain.
The second result obtained after ARM processes exports can to the data transfer mode between GPU, ARM and GPU
Thinking: ARM sends data to the buffer queue pre-build, GPU is from the buffer queue pre-build asynchronous acquisition data.Base
In aforementioned data transfer mode, each step of ARM, GPU can completely asynchronous perform, and couples the lowest so that extends through
GPU number, ARM core number, that buffer queue length improves computing capability is simple.
Owing to, in the low-resolution image decoding little at parallel scale, high frequency sharpens the GPU of link and processes and be no faster than string
The ARM process of row, but the GPU processing speed that high frequency sharpens in the decoding of high-definition picture link is close with ARM, will
Distribution of computation tasks, to GPU, therefore, it can the resolution value according to image, and on the premise of not affecting decoding speed, determining will
High frequency sharpens link and is dispatched on GPU, and it is sharp that the second result that ARM is carried by GPU carries out high frequency according to high frequency sharpening algorithm
Change processes.
GPU can use a special daemon thread, be responsible for video memory-internal storage data transmission and with other threads
Communication, in order to know and read and write whether can using of purpose buffering.Aforementioned special daemon thread is responsible in the middle of ARM-GPU
Data queue fetches data, and uploads to video memory, the processing stage of performing GPU, result descends into the correspondence position of output buffering array
In, it is responsible for communicating with the ARM thread of other links simultaneously.ARM scanning thread obtains the queue not empty signal of GPU daemon thread
After, run-down output buffer queue, by data output minimum for index.
On the one hand calculating the computational threads used can expand as required, the number of the GPU kernel of use can also
Being set to two or more, buffer length arbitrarily can also adjust according to internal memory, so expands computing capability simultaneously and handles up to increase
Rate is the simplest, it is also possible to the self-defined adaptive tuning to hardware such as ARM, GPU, internal memories；On the other hand inside is achieved
The most out of order concurrently decoding, and outside program, the sequence order of output is identical with list entries, output keeps original
Sequentially.
Can be high (between codeblock) according to view data degree of parallelism in coarseness, complete serial in fine granularity
The spy of a kind of instruction once can only be performed inside the feature of (codeblock self), and mono-Local size sets of threads of GPU
Point, works out the mapping relations between GPU thread and data.Particularly as follows: the second result is being entered according to high frequency sharpening algorithm
During row Edge contrast, as it is shown in fig. 7, the second result is divided into several data blocks, the GPU calling electronic equipment is
Each video data block one thread bundle Local size of distribution in two results, data block and thread bundle one_to_one corresponding,
The corresponding different thread bundle Local size of different images data block；With Local size for processing unit to the second result
In each video data block carry out high frequency sharpening algorithm process, obtain the 3rd result.
Concrete, in units of data block, carry out data process, independent calculating between data block.Inside data block, do not have
There are strong iterative and dependency, it is easy to parallel.Image more long data block total amount is the most.Therefore parallel mode is: 1, between data block
Concurrently, the corresponding thread bundle of each data block；2, the thread in OpenCL is with Local size (general number is for 16) for single
Position is organized and dispatches；3, a Local size once can only perform the branch in a calculating, in high frequency sharpening algorithm
Portion is almost full with branch, and therefore preferably way is that every 16 threads only use one to perform calculating, saves data transfer bandwidth；
4, the element branches needed for the coding that counts is judged that operation is converted to table lookup operations；5, shared drive storage intermediate object program is used.
It addition, in the present embodiment, it is a data block by each pixel definition in the second result, then, the thread of distribution
The number of bundle is just identical with the number of pixel, and is relation one to one between thread bundle and pixel.
In the case of being responsible for by GPU performing the subsequent treatment to the second result, ARM can be by the first result
Send to video memory, in order to GPU quick obtaining data process.The parallel ability utilizing GPU directly will sharpen it in video memory
After the result (the 3rd result) that obtains carry out local histogram's blending algorithm process with the first result, to improve
GPU performs the efficiency of memory read data.Utilize video memory caching and the exchanging visit data mapping mechanism of ARM memory cache of GPU, with
Reduce from the interior data copy number of times being stored to video memory, the data reading speed of optimization.
The GPU of electronic equipment performs high frequency and sharpens and local histogram's fusion, as shown in Figure 8, may include that this module is defeated
Enter 16 bit data linear stretch view data, wherein 10 effectively, input 8 bit data partial equilibriums and strengthen data.Algorithm is at GPU
Middle executed in parallel, once calculates 2 float pixel datas, needs 320 × 512 threads, the global of each GPU kernel
Size distributes 320 × 512 threads, and local size distributes 8 × 8 threads, does not has dependency between thread-data, and module is real
Existing flow process is as shown in Figure 8.In order to reduce GPU memory access, therefore the present invention uses vector once to capture two pixels in view data
Float2, and make vector operation, in order to reduce the compiling instruction number of GPU kernel function.
As can be seen here, based on ARM and GPU the infrared DDE method of the present invention, call in the decoding of high-definition picture
The speed that GPU performs high frequency sharpening algorithm is close with ARM, the execution of high frequency sharpening algorithm is operated and is assigned to GPU, on the one hand exists
The load of ARM is alleviated on the premise of not affecting decoding speed, the idle GPU resource on the other hand made full use of, thus real
Show the load balancing of ARM, GPU, increase the overall throughput of system.
It addition, GPU can carry out the parallelization uploaded, perform, download during performing.Concrete, use GPU's
Stream characteristic, when performing present image decoding, completes the result of previous frame image is transferred to internal memory, and in advance by next
The digital independent of two field picture to caching in, conceal data transmission time.
GPU interface in OpenCL definition is broadly divided into three classes: data are uploaded, performed parallel computation, in downloading the data to
Depositing, three kinds of operations can be asynchronous, but correctly to obtain result, be in correct time synchronized.This three class must serial
Perform, logical error otherwise can occur.OpenCL defines new structured data stream stream it can be understood as upload/perform/
The container of download sequence.In OpenCL, the operation with type cannot be parallel, uploads behaviour between the most multiple data stream stream
Work can only be serial.Therefore the present invention proposes parallel mode based on stream as shown in Figure 9, and in figure, transverse axis is operation
The actual execution time, the longitudinal axis of inclination is the triggered time, namely the time of call operation interface.One anti-" L " type square frame
Be inside once circulation in execution content, it can be seen that use as shown in Figure 9 asynchronous API Calls order, upload to video memory, under
Being downloaded to internal memory and perform operation time by overlapping well, the actual used time is reduced to the longest that in three by three's sum
Individual, significantly reduce overall time.
It is also preferred that the left for the decoding scheme of ARM and GPU isomerism parallel frame by frame.Concrete, by the view data of Real-time Collection
Frame by frame its incoming ARM, GPU being realized line production according to degree of parallelism, such hardware computing resource can fully be used, no
There will be the free time, improve overall data throughput.
Offer ARM-GPU load regulation mechanism is provided, gives full play to hardware performance, increase feature of image
Adaptability, various images can efficiently process.User need not be concerned about the details of each algorithm or understand the mapping of hardware algorithm
Mode, it is only necessary to select suitable passage can obtain preliminary optimization according to feature of image and limiting of calculating resource.
In order to verify the superiority of the inventive method further, illustrate below by the data in following form.
According to the link that infrared DDE is main: the stretching of filter preprocessing, gray scale, restriction contrast equalize, header resolves,
Local histogram's bilinear interpolation and high frequency sharpen and local histogram merges, and above-mentioned decoding link is used ARM, GPU respectively
Realize.The mapping relations of each decoding link are as shown in table 1 below:
Table 1
By table can be seen that, DDE link has different computing features, be suitable for parallel module can parallel form the most not
The most identical.By can be parallel part use suitable mode to be mapped to the thread resources of GPU, make full use of GPU computing unit many,
The advantage that degree of parallelism is high, evades GPU and is bad to process the weakness of the computing that branch is many, dependency is high, make full use of calculating resource.
Needing internal storage data to upload to video memory when ARM calculation stages is transitioned into GPU, in the GPU computing interval, data reside in aobvious always
Depositing, change thread mapping mode and need not the extra time, demonstrating the inventive method is optimum in theory.Real in reality
In testing, such as, it is 16bit bit depth single channel in input picture parameter, uses IMX6 platform, process 25fps per second can be reached
Process data.As can be seen here, the inventive method makes full use of the computing capability of current computer multinuclear ARM+GPU framework, also
Improve the throughput of real time image collection parallel processing.
In the present embodiment, the Processing Algorithm related in each step above-mentioned is existing conventional algorithm, so, the most not
Again each algorithm itself is described in detail.
In above-described embodiment, each step in image detail enhancement method is illustrated, and is given
Corresponding detailed description of the invention, but, the basic ideas of the present invention are above-mentioned basic technical scheme, so, the present invention is not
It is confined to the above-mentioned specific descriptions for each step.
Device embodiment
Corresponding to method mentioned above embodiment, present invention also offers a kind of infrared enhancing based on SOC framework dress
Put.
This device is specially configured with the embedded SOC framework electronics of central processing unit ARM and graphic process unit GPU and sets
Standby, in this electronic equipment:
Central processing unit ARM performs following software module:
Filtering Processing module, is filtered according to filtering budget law for the view data receiving data reception module
Processing, Filtering Processing is divided into dark order to carry out mean filter and medium filtering pretreatment；
Gradation processing module, for carrying out 14 to 10 to the view data after filtered process according to gray scale stretching algorithm
The position linear stretch processing of gray scale；
Contrast processing module, for carrying out according to limiting contrast equalization algorithm the view data after gray proces
Contrast processes；
Rectangular histogram interpolation processing module, the view data after processing degree by contrast is according to local histogram's bilinearity
Interpolation algorithm carries out rectangular histogram interpolation processing；
The graphic process unit GPU following software module of execution:
Edge contrast module, for carrying out sharp to the view data after rectangular histogram interpolation processing according to high frequency sharpening algorithm
Change processes；
Rectangular histogram Fusion Module, for by the view data after gray proces be sharpened process after view data enter
Row carries out fusion treatment according to local histogram's blending algorithm, and the view data after process is i.e. to obtain image detail enhancement process
After infrared image.
As can be seen here, the image detail intensifier based on ARM and GPU of the present invention, in the decoding of high-definition picture
In call GPU to perform the speed of high frequency sharpening algorithm close with ARM, the execution of high frequency sharpening algorithm operation is assigned to GPU, one
Aspect alleviates the load of ARM on the premise of not affecting processing speed, the idle GPU resource on the other hand made full use of,
It is achieved thereby that the load balancing of ARM, GPU, increase the overall throughput of system.
Further, rectangular histogram Fusion Module is data int to float format converting module, and for utilizing, GPU's is parallel
Ability directly in video memory by the view data after gray proces be sharpened process after view data carry out local Nogata
Figure blending algorithm processes, to improve the efficiency of the GPU execution memory read data of electronic equipment.
For device embodiment, owing to the modules in this device is functional module, it is by accordingly
Software program realizes corresponding function, so, this device is the most still method, below central processing unit ARM execution
Operation: the view data receiving data reception module is filtered processing according to filtering budget law, and Filtering Processing is divided into secretly
Order carries out mean filter and medium filtering pretreatment；View data after filtered process is carried out according to gray scale stretching algorithm
14 to 10 the linear stretch processings of gray scale；View data after gray proces is carried out according to limiting contrast equalization algorithm
Contrast processes；View data after processing degree by contrast carries out rectangular histogram according to local histogram's bilinear interpolation algorithm and inserts
Value processes；
Following operation is performed: sharp according to high frequency to the view data after rectangular histogram interpolation processing by graphic process unit GPU
Change algorithm and be sharpened process；By the view data after gray proces be sharpened process after view data carry out according to office
Portion's rectangular histogram blending algorithm carries out fusion treatment, and the view data after process is i.e. infrared after obtaining image detail enhancement process
Image.
Owing to the method has been carried out detailed description in said method embodiment, so described in the device embodiment
Fairly simple, relevant part sees the respective description of embodiment of the method.
It is presented above specific embodiment, but the present invention is not limited to described embodiment.The base of the present invention
This thinking is above-mentioned basic scheme, for those of ordinary skill in the art, according to the teachings of the present invention, designs various change
The model of shape, formula, parameter are not required to spend creative work.The most right
Change that embodiment is carried out, revise, replace and modification still falls within protection scope of the present invention.
Claims (6)
1. an infrared image detail enhancing method, it is characterised in that utilizing SOC framework to process infrared image, this is red
Outer image detail enhancement method is:
The ARM in SOC framework is utilized to carry out following operation:
It is filtered processing and carrying out gray proces according to gray scale stretching algorithm according to filtering budget law successively to view data,
Image is processed to first；
Process image to described first and carry out contrast process and according to local Nogata according to restriction contrast equalization algorithm successively
Figure bilinear interpolation algorithm carries out rectangular histogram interpolation processing, obtains the second process image；
The GPU in SOC framework is utilized to carry out following operation:
Process image to described second and be sharpened process according to high frequency sharpening algorithm, obtain the 3rd process image；By described
Three process images and first process image and carry out fusion treatment according to local histogram's blending algorithm, and the view data after process is i.e.
It it is the infrared image after obtaining image detail enhancement process.
Infrared image detail enhancing method the most according to claim 1, it is characterised in that the data between ARM and GPU pass
Defeated mode is: ARM sends data to the buffer queue pre-build, GPU obtains from the described buffer queue pre-build is asynchronous
Fetch data.
Infrared image detail enhancing method the most according to claim 1, it is characterised in that realize processing figure to described second
As being sharpened the means of process according to high frequency sharpening algorithm it is: the second process image is divided into several image blocks, and for institute
Stating the thread bundle that the second process image distribution is identical with described image block number, image block and thread bundle one_to_one corresponding, with thread
Bundle is sharpened process for processing unit to the described second each image block processing image.
4. an infrared image details intensifier, it is characterised in that include ARM and GPU,
Described ARM performs following processing module:
Filtering Processing module, for being filtered processing according to filtering budget law to view data；
Gradation processing module, for carrying out at gray scale according to gray scale stretching algorithm the view data after described Filtering Processing
Reason；
Contrast processing module, for carrying out according to limiting contrast equalization algorithm the view data after described gray proces
Contrast processes；
Rectangular histogram interpolation processing module, the view data after processing degree by contrast is according to local histogram's bilinear interpolation
Algorithm carries out rectangular histogram interpolation processing；
Described GPU performs following processing module:
Edge contrast module, for carrying out sharp to the view data after described rectangular histogram interpolation processing according to high frequency sharpening algorithm
Change processes；
Rectangular histogram Fusion Module, for by the view data after described gray proces and the picture number after described Edge contrast
Carrying out fusion treatment according to local histogram's blending algorithm according to carrying out, the view data after process is i.e. to obtain image detail to strengthen
Infrared image after process.
Infrared image details intensifier the most according to claim 4, it is characterised in that the data between ARM and GPU pass
Defeated mode is: ARM sends data to the buffer queue pre-build, GPU obtains from the described buffer queue pre-build is asynchronous
Fetch data.
Infrared image details intensifier the most according to claim 4, it is characterised in that described Edge contrast module realizes
The means that view data after described gray proces is sharpened process are: by the view data after described gray proces
It is divided into several image blocks, for the line that the distribution of described view data after described gray proces is identical with described image block number
Cheng Shu, image block and thread bundle one_to_one corresponding, with thread bundle for processing unit to described picture number after described gray proces
According to each image block be sharpened process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610684865.3A CN106296616B (en) | 2016-08-18 | 2016-08-18 | A kind of infrared image detail enhancing method and a kind of infrared image details enhancement device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610684865.3A CN106296616B (en) | 2016-08-18 | 2016-08-18 | A kind of infrared image detail enhancing method and a kind of infrared image details enhancement device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106296616A true CN106296616A (en) | 2017-01-04 |
| CN106296616B CN106296616B (en) | 2019-01-29 |
Family
ID=57679278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610684865.3A Active CN106296616B (en) | 2016-08-18 | 2016-08-18 | A kind of infrared image detail enhancing method and a kind of infrared image details enhancement device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106296616B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107240081A (en) * | 2017-06-20 | 2017-10-10 | 长光卫星技术有限公司 | The denoising of night scene image and enhancing processing method |
| CN107451986A (en) * | 2017-08-10 | 2017-12-08 | 南京信息职业技术学院 | A kind of single width infrared image enhancing method based on integration technology |
| CN110827229A (en) * | 2019-11-01 | 2020-02-21 | 中国科学院长春光学精密机械与物理研究所 | Infrared image enhancement method based on texture weighted histogram equalization |
| CN110895788A (en) * | 2019-11-29 | 2020-03-20 | 上海众链科技有限公司 | System for enhancing graphic processing capability and external device |
| CN111897639A (en) * | 2020-07-29 | 2020-11-06 | 平安科技（深圳）有限公司 | Image augmentation method, image augmentation device, computer device, and storage medium |
| CN111970564A (en) * | 2020-08-26 | 2020-11-20 | 展讯通信（上海）有限公司 | Optimization method and device for HDR video display processing, storage medium and terminal |
| CN112532856A (en) * | 2019-09-17 | 2021-03-19 | 中兴通讯股份有限公司 | Shooting method, device and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101153850A (en) * | 2006-09-30 | 2008-04-02 | 长安大学 | Method and system for detecting asphalt mixture |
| CN102801912A (en) * | 2012-07-17 | 2012-11-28 | 北京伽略电子***技术有限公司 | Infrared image processing chip and processing method |
| CN103544470A (en) * | 2013-08-05 | 2014-01-29 | 华中科技大学 | Double-color infrared isomerism parallel automatic air target identifier for movable platform |
| CN104866286A (en) * | 2015-06-02 | 2015-08-26 | 电子科技大学 | OpenCL and SoC-FPGA-Based K neighbor sorting accelerating method |
| US9224362B2 (en) * | 2013-03-14 | 2015-12-29 | Microsoft Technology Licensing, Llc | Monochromatic edge geometry reconstruction through achromatic guidance |
-
2016
- 2016-08-18 CN CN201610684865.3A patent/CN106296616B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101153850A (en) * | 2006-09-30 | 2008-04-02 | 长安大学 | Method and system for detecting asphalt mixture |
| CN102801912A (en) * | 2012-07-17 | 2012-11-28 | 北京伽略电子***技术有限公司 | Infrared image processing chip and processing method |
| US9224362B2 (en) * | 2013-03-14 | 2015-12-29 | Microsoft Technology Licensing, Llc | Monochromatic edge geometry reconstruction through achromatic guidance |
| CN103544470A (en) * | 2013-08-05 | 2014-01-29 | 华中科技大学 | Double-color infrared isomerism parallel automatic air target identifier for movable platform |
| CN104866286A (en) * | 2015-06-02 | 2015-08-26 | 电子科技大学 | OpenCL and SoC-FPGA-Based K neighbor sorting accelerating method |
Non-Patent Citations (1)
| Title |
|---|
| 孔颖军: "基于红外图像的道路识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107240081A (en) * | 2017-06-20 | 2017-10-10 | 长光卫星技术有限公司 | The denoising of night scene image and enhancing processing method |
| CN107451986A (en) * | 2017-08-10 | 2017-12-08 | 南京信息职业技术学院 | A kind of single width infrared image enhancing method based on integration technology |
| CN112532856A (en) * | 2019-09-17 | 2021-03-19 | 中兴通讯股份有限公司 | Shooting method, device and system |
| CN112532856B (en) * | 2019-09-17 | 2023-10-17 | 中兴通讯股份有限公司 | Shooting method, device and system |
| CN110827229A (en) * | 2019-11-01 | 2020-02-21 | 中国科学院长春光学精密机械与物理研究所 | Infrared image enhancement method based on texture weighted histogram equalization |
| CN110895788A (en) * | 2019-11-29 | 2020-03-20 | 上海众链科技有限公司 | System for enhancing graphic processing capability and external device |
| CN111897639A (en) * | 2020-07-29 | 2020-11-06 | 平安科技（深圳）有限公司 | Image augmentation method, image augmentation device, computer device, and storage medium |
| CN111970564A (en) * | 2020-08-26 | 2020-11-20 | 展讯通信（上海）有限公司 | Optimization method and device for HDR video display processing, storage medium and terminal |
| CN111970564B (en) * | 2020-08-26 | 2023-03-24 | 展讯通信（上海）有限公司 | Optimization method and device for HDR video display processing, storage medium and terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106296616B (en) | 2019-01-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106296616A (en) | A kind of infrared image detail enhancing method and a kind of infrared image details intensifier | |
| CN109472270B (en) | Image style conversion method, device and equipment | |
| CN108898567B (en) | Image noise reduction method, device and system | |
| CN109064396B (en) | Single image super-resolution reconstruction method based on deep component learning network | |
| CN104685538B (en) | System and method for reducing the noise in video flowing | |
| WO2020199831A1 (en) | Method for training image processing model, image processing method, network device, and storage medium | |
| JP2023504669A (en) | Image processing method, smart device and computer program | |
| US20210233210A1 (en) | Method and system of real-time super-resolution image processing | |
| CN106663314A (en) | Real time skin smoothing image enhancement filter | |
| GB2471647A (en) | Object Detection Using Feature Responses | |
| CN108875900A (en) | Method of video image processing and device, neural network training method, storage medium | |
| CN112966635B (en) | Low-resolution time sequence remote sensing image-oriented moving ship detection method and device | |
| CN112070664A (en) | Image processing method and device | |
| CN110428382A (en) | A kind of efficient video Enhancement Method, device and storage medium for mobile terminal | |
| CN116547694A (en) | Method and system for deblurring blurred images | |
| CN110717864B (en) | Image enhancement method, device, terminal equipment and computer readable medium | |
| CN112470472B (en) | Blind compression sampling method and device and imaging system | |
| CN104376546A (en) | Method for achieving three-path image pyramid fusion algorithm based on DM642 | |
| Hung et al. | Video super resolution via deep global-aware network | |
| CN109345464A (en) | A kind of method and system of image procossing that realizing HDR in Bayer data field | |
| CN106210727B (en) | Video hierarchical code stream coding method and system based on neural network processor array | |
| CN104954749B (en) | A kind of information recording method | |
| CN105959512B (en) | 3D noise reduction implementation method and device | |
| CN111461957A (en) | Reconfigurable design image processing module | |
| CN104883573B (en) | A kind of signal high-efficient treatment method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |