Nevertheless, contemporary and most sensor all uses CMOS unit, this basically is the consideration that stems from production respect. Sensor and optical parts of an apparatus often conformity is used at making chip class camera together, this kind of camera is used to be in be similar to biology or microscope to learn to wait for a domain, if pursue,1 is shown.

 

Graph 1: integrated of the image sensor of optical parts of an apparatus and color filter commonly used arrange

 

Image sensor is to satisfy the special target of different application design, it offerred the sensitivity of different level and quality. Want to be familiar with all sorts of sensor, can consult its firm information. For example, to be in silicon radical model and trends are answered (use at realizing luminous intensity and color to detect) between have best eclectic, specific to semiconductor makes a process, need to optimize the size of unit of sensor of every photoelectricity diode and compositive composition.

 

To computer vision character, the effect of sampling theory has important sense, if the limits resembling element of target setting is met,use Nyquist frequency. Sensor resolution and optical parts of an apparatus can resemble element offerring enough resolution for every together, so that undertake resembling to be interested in a feature, because this has such verdict: The sampling of interest feature (or into resemble) frequency should be important like element (to interested feature) in most double of figurine element size. Of course, to becoming like precision word, of double crossing sampling is target of a lowest merely, in applying actually, decide sheet resembles the feature of element width not easily.

 

To given applying, want to obtain best result, want calibration camera system, so that be below different illumination and distance condition,resemble element certainly deepness (Bit Depth) sensor noise and dynamic limits. The noise that produces to any color passageway place to can handle sensor and nonlinear answer, and detect build a model lack fidelitily with geometry of bad place of element of school positive-appearing image, processing, need to develop appropriate sensor processing technique. If use test mode to devise method of a simple demarcate, this kind of method resembles the respect such as element size having in gray scale, color, feature by fine to thick gradual change, can see a result.

 

1, sensor material

 

Silicon cartography is the widest like sensor application, also can use other data of course, the meeting in applying with the military affairs in industry for instance uses gallium (Ga) will enclothe longer than silicon infra-red wavelengh. Different camera, the resolution of its image sensor can differ somewhat. Resemble camera of element photoelectricity transistor from sheet (it scans point-blank through one dimension array is used at industrial application) , the array of 2 dimension rectangle to common camera (all going to globose the method of entire line all is used at high resolution to become picture) , use likely. (this chapter can introduce sensor configuration and camera configuration finally) .

 

Common into use CCD, CMOS, BSI and Foveon method to undertake production like sensor. Silicon cartography has a nonlinear spectrum to answer a curve like sensor, this meeting feels spectral close infra-red part well, but mix to blue, purple close violet exterior cent is felt badly (if pursue,2 are shown) .

 

Graph 2: The typical spectrum of diode of photoelectricity of a few kinds of silicon is answered. Can notice, photoelectricity diode has tall susceptibility inside the close infra-red limits around 900 accept rice, and in over or across ～ of 400 accept rice has nonlinear susceptibility inside the visible light limits of 700 accept rice. As a result of the cause that standard silicon answers, from inside camera filter of take out IR can increase close infra-red sensitivity. (the license that the use of spectral data image already obtained OSI photoelectricity Inc. )

 

Attention, should read in primitive data, when changing this data disperse the number to resemble element, can bring about silicon spectrum to answer. Sensor manufacturer did design compensation in this area, however, when the basis uses demarcate camera system and devising sensor processing technique, the facial expression that should consider sensor is answered.

 

2, component of sensor photoelectricity diode

 

The key of image sensor depends on the size of the size of photoelectricity diode or component. The photon amount that the sensor component place that uses small photoelectricity diode captures did not use big photoelectricity diode much. If component dimension is less than the visible light wavelengh that can capture (the La Guang that is 400 accept rice like length) , so resemble color for correction chart, other issue must be overcome in sensor design. Sensor manufacturer costs a large number of energy to design optimize component size, become in order to ensure all color can be coequal picture (if pursue,3 are shown) . Below extreme circumstance, because lack accumulative photon and sensor,numerate noise, small sensor may be more sensitive to noise. If 2 extremely element of glow valve sensor is too large, so the grain size of silicon material and cost can raise, this does not have any advantages but character. The sensor component volume that equipment of general business sensor has is 1 square micron at least, every manufacturer will be different, but can have a few to satisfy certain and special requirement eclectic.

 

Graph 3: The wavelengh of primary color allocates. Attention, primary color area overlaps each other, to all color character, green is a very good monochromatic substitute

 

3, sensor configuration: Mosaic, Faveon and BSI

 

Graph 4 those who showed much more spectral sensor is designed is different piece inside configuration, include mosaic and pile method. In mosaic method, color filter is installed to be on the mosaic mode of every component. Method of Faveon sensor pile relies on color wavelengh deepness to permeate the physical composition of semiconductor material, among them the osmosis that every kinds of color has different rate to silicon material, undertake resembling to respective color thereby. Whole component size applicably at all color, do not need to be every kinds of color to configure component respectively so.

 

Graph 4: (Zun Tu) the Foveon method of pile RGB component: There is RGB color in every component position, absorb different wavelengh in different deepness; (right graph) mosaic component of the standard: In every photoelectricity filter of a RGB is placed above diode, every filter allows specific wavelengh to cross every photoelectricity diode only

 

Retrorse illume (Back-side Illuminated, BSI) sensor structure has larger component area, and every cell should assemble more photon, decorated sensor wiring afresh on grain consequently.

 

Of sensor component decorate also affect color to answer. For example, graph 5 showed primary color (R, G, B) of sensor and white sensor different arrange, among them white sensor (W) the color filter that has a special clarity or blame color. The permutation of sensor considered the processing resembling element of certain limits, if be in the processing process that sensor resembles element information to, what choose in the different configuration that can combine element of in the vicinity of resemble element, these information resembling element can optimize color to answer or dimensional color resolution. Actually, certain application uses primitive sensor data to carry out common processing procedure to enhance resolution merely mixture of tectonic perhaps and other facial expression.

 

 

Graph 5: A few of component color different mosaic are configured, include color of white, basic RGB and component of less important CYM. Every kinds of configuration handles a process to optimize color for sensor or dimensional resolution provided different means (image comes from at ” Building Intelligent Systems ” one book, the use that gets Intel publishing house is permitted) .

 

The size of whole sensor also decided the size of camera lens. Generally speaking, the light that camera lens passes greatly more is more, accordingly, to photography application character, bigger sensor can apply to digital camera better. Additional, the aspect ratio that component arranges on grain (Aspect Ratio) the geometrical form that decided to resemble element, be like, 4: 3 with 3:2aspect ratio is used at digital camera and the film of 35 millimeter respectively. The detail that sensor configures is worth a reader to understand, such ability design best sensor handle process and image preprocesser quite.

 

4, dynamic limits and noise

 

Current, every color unit can offer the most advanced sensor at least 8 bit, it is 12 ～ normally 14 bit. Sensor component needs to cost space and time to collect photon, so lesser component must pass elaborate design, produce a few problems in order to avoid. Component of the optical component that the likelihood comes from noise to be used at place, color filter, sensor, gain and A/D converter, later period handles a process to perhaps compress a method to wait. Of sensor numerate noise also can affect actual resolution, because every resemble element unit numerating from inside sensor,pass in A/D converter again, form the travel of digital form and line thereby, so that be used at resembling element changeover. Better sensor can produce less noise, can get more efficient bit resolution at the same time. The job of Ibenthal is to fall the fine literature of respect of a confusion of voices.

 

Additional, sensor photon is absorbed can differ somewhat to every kinds of color, have some of problem likely to blue, become like word to lesser sensor namely this is a kind of the most difficult color. Below certain circumstance, manufacturing chamber of commerce tries to be every kinds of color in sensor built-in a simple gamma curve revises a method, but this kind of method is not worth to advocate. In the application that has demand to color, can consider chroma equipment model and color government, the passage of every kinds of color that lets sensor even is had nonlinear the feature is built a series of simple corrective search a watch (Lookup Table, LUT) changeover.

 

5, sensor processing

 

Sensor processing is used at mosaic goes in array and gathering from sensor to resemble element, also use at corrective perception flaw. In this one our meeting discussion sensor handles a foundation.

 

Become in every normally there is processor of sensor of a proprietary in resembling a system, include interface of sensor of a fast HW, optimized train in excess specified length instruction collect (Very Long Instruction Word, VLIW) , only statement much data flows (Single Instruction Multiple Data, SIMD) instruction and the hardware module that have fixed function, these functions are to solve the working load that large-scale collateral processing resembling element causes. Normally, sensor processing process is transparent and automation, and by install into the manufacturer that resembles a system, come from all image of sensor all be handled with same way. Also put in the other way that is used at providing primitive data, these data allow needle correspondence to use custom-built sensor to handle a process, photograph like the number in that way.

 

6, go mosaic

 

Configure according to different sensor element (like the graph 5 are shown) , can use all sorts of going to mosaic algorithm generates primitive sensor data final RGB to resemble element. Losson &Yang still had the person such as Li to give out respectively two first-rate overview document, these document introduced all sorts of methods and the challenge that face to wait.

 

Go one of a of mosaic main challenges are interpolation resembling element, the color passageway that its action is will adjacent unit is combined individual like element. Below the condition that gives the aspect ratio that the geometrical appearance that sensor component arranges and unit arrange, this is a serious problem. To it the problem increasing advantageous position that pertinent question is color unit, if be in element of every RGB resembling,every kinds of color should hold how many proportion. Because be in mosaic sensor, dimensional component resolution is more than the RGB that combines finally to resemble element resolution, certain application needs primitive sensor data, so that use all precision and resolution as far as possible, perhaps some of processing or needs to enhance efficient resolution resembling element, or need realizes a space to accurate color is handled and go better mosaic processing.

 

7, bad those who resemble element is corrective

 

Like LCD monitor, sensor also may have bad like element. Want corrective bad coordinate resembling element through supply and demand is being lifted in camera module or driver, the supplier can be in a factory corrective sensor, offer plan of blemish of a sensor for foregone blemish. Below certain circumstance, adaptive blemish method of correction can be used on sensor, so that monitoring vicinity resembles element dot discover flaw, next the blemish type inside corrective and certain limits, for instance sheet resembles the lump blemish of row or element blemish, travel blemish and similar 2×2 or 3×3. Search a flaw for real time, camera drive also can be offerred adaptive blemish analysis, a special compensation control may be offerred in the starting menu of camera.

 

8, color and illume are corrective

 

Be necessary to have color corrective so that balance total color accuracy and Bai Pingheng. If graph 1-2 place is shown, mix to red on silicon sensor green these two kinds of color normally very sensitive, but not sensitive however to blue, accordingly, understanding and demarcate sensor are the main job that gets the most accurate color.

 

The processor of image sensor included great majority to use Yu Guangyun’s corrective geometry processor, this is illumination in the brim expression of image darker. Corrective be based on geometrical twist function, can consider programmable illumination function to increase the illumination of front brim, this need undertakes demarcate before leave factory, so that match with optical halation mode photograph.

 

9, geometry is corrective

 

Camera lens may have geometry to differ or face brim produces twist, produce radial and lack fidelity picture. To solve camera lens distortion, great majority is become have appropriative sensor processor like the system, it has the digital twist element that a hardware quickens, be similar to the grain sampler on GPU. The geometry that can be aimed at optical parts of an apparatus in the factory is corrective undertake calibration and process designing.