TW200805178A - Monolithic image perception device and method - Google Patents

Abstract

An apparatus which can acquire, readout and perceive a scene based on the insertion, or etching of photosensitive elements into or on a transparent or semi-transparent substrate such as glass. The substrate itself acts as the optical device which deflects the photons incident to the reflected image into the photosensitive elements. Photosensitive elements are interconnected together by a transparent or opaque wiring. A digital neural memory can be trained to recognize specific scenery such as a human face, an incoming object, a surface defect, rain drops on a windshield and more. Other applications include image-perceptive car headlight and flat panel display detecting and identifying the viewer's behavior (gaze tracking, face recognition, facial expression recognition and more). Yet another application includes sliding doors perceiving the direction and speed of an individual coming towards that door. Yet another application includes permanent damage detection (texture change) in dam, bridge or other manmade construction.

Description

200805178 IX. Description of the Invention: TECHNICAL FIELD OF THE INVENTION The present invention generally relates to an image forming apparatus. In particular, the present invention relates to a microdevice for image recognition disposed on a transparent substrate such as glass or embedded in a transparent substrate such as glass. [Prior Art] Transparent surfaces such as glass have existed for hundreds of years. The transparent surface was originally intended to protect the living space while giving the occupants a sense of the outside world (landscape, weather and possible threats). More recently, a large number of transparent surfaces have been required for display industries, starting with cathode ray tubes (CRTs) and more recently for liquid crystal displays (LCDs) and many other types of flat panel displays. In use, in most cases, humans or living organisms (animals, plants) are located near these transparent surfaces. Image sensors have been in use for decades (for example, CCD or CM〇s sensors). For example, see U.S. Patent No. 6,617,565 for CMOS image sensors, the contents of which are incorporated herein by reference. A typical image sensor is based on a camera design and generally includes an integrated circuit located behind the lens, which lens can be micro or movable (e.g., a helical mounting lens). The sensor is used to convert light energy (photons) into electrical signals that are proportional to the light-emitting elements of the light-sensitive elements that are organized in an array on the sensor. The image is synthesized based on the output of the photosensitive element. Image recognition technology has become a boon. Cameras of all sizes and manufactured are needed for applications such as safety, identification, intelligence, quality inspection, traffic surveillance and more. The camera is usually linked to the display device by a wired or wireless connection. Today, mobile phones are conventionally equipped with a miniature camera connected to the LCD display device in which 112868.doc 200805178 is placed. Advanced image recognition requires high-resolution imaging synthesis. Due to insufficient processing power and/or because the processor can only process the pixels of the image, b nowadays image recognition systems operate at relatively low speeds. Therefore, there is a need for new imaging improvements that have been improved in the prior art. SUMMARY OF THE INVENTION The object of the present invention is to provide an image recognition device that has a sensing region (for example, a photosensitive member) directly included in a transparent or translucent material constituting an optical interface between a human image and a sensing region. ). The image recognition device itself is preferably transparent or translucent. Yet another object of the present invention is to provide a sensing region having a region's decision making capability by means of an array of trainable processing elements. In one embodiment of the invention, the trainable cognitive memory component or unit is associated with one or more photosensitive elements The regional decision making capability provides the advantage of reducing the transmission requirements (i.e., bandwidth) of the device, especially when the number of photosensitive elements is large and when the transmission frequency of the photosensitive elements must be high. A high-resolution, high-speed imaging device can be obtained with a large array of sensing regions of regional decision-making capabilities. The trainable cognitive memory component can be operated at a low frequency and causes very low according to an embodiment of the present invention. The current, because &, ensures that each element is self-disciplined and can use very economical energy sources (such as too uniform). According to an embodiment of the invention, one or more of the substrates are lost to one or more 112868. Doc 200805178 Photosensitive elements are associated with one or more novel single-beam, beam-like elements to form a new early stone image recognition device. According to the invention, a plurality of photosensitive elements of the game # ytu can be used; the training cognitive elements can be associated with transparency or rich reading | 4 j operation configuration and spread over the flatness. Connection column ^ (four) is not limited to linear array of parallel neurons

/ Array of neurons connected in a rectangular matrix or a honeycomb geometry. Further applications and advantages of various embodiments of the present invention I 5 are discussed below with reference to the drawings. [Embodiment] While the invention may be embodied in a number of different forms, the invention is described herein, and is not to be construed Any particular preferred embodiment described and/or illustrated herein.

The present invention is an imaging device that can include a sensor sensing device such as a photosensitive element that is coupled to, can be combined with, or otherwise associated with a trainable cognitive element (bah讣 & cognitive element) Training cognitive elements are associated, wherein two elements are chemically deposited or otherwise deposited on the surface of the transparent substrate or embedded in the surface of the transparent substrate. Throughout this document, the combination of a sensing area and a trainable cognitive element with "regional" decision-making ability is called a "cognitive sensor." Throughout this document, the training cognitive element is called "CogniMem". The sensing area is generally controlled by one or Multiple photosensitive elements are constructed, but may cover other sensing configurations. 112868.doc 200805178 According to this gx example, the cognitive sensor can be configured to recognize an incident light pattern (eg, an image or image portion), and to process an incident light pattern. In order to generate regional decision-making and transmission area decision-making instructions, cognitive sensors may include, for example, (but not limited to) regional decision making capabilities _ data input logic neurons and decision output logic, memory buffers , solar cells for energy auton〇 (energy auton〇my) and more. Each cognitive sensor preferably characterizes a reactive learning memory (reactive _ aSS〇Clatlve learning memory; REALM) in parallel configuration. According to the actual embodiment of the present invention, CogniMem is capable of digital or analog pattern recognition without any computer instructions.

CogmMem may comprise one or more neurons, which are related memories that are accessible in parallel, which are responsive to input patterns similar to their own content. Neurons can respond individually or collectively by activating their responses based on responses from other neighboring neurons. This selection can be done via a suppression/excitation input line connected to the neuron. • The neuron content of C〇gniMem constitutes "knowledge". Knowledge is a set of static distinct digital signatures. Knowledge can be static knowledge (one-time load) or dynamic knowledge ^ (updated by other neurons' responses or properly loaded from an external knowledge base), but is preferably automatically generated by the learning process (learning pr〇cess) I need a computer to do this. Use the same or different knowledge of deposition on the same substrate.

The CogniMem can be deposited on the substrate or embedded in the substrate (or otherwise coupled to the substrate) as part of the cognitive sensor, or as a separate part. In the former case, CogniMem is usually used to identify the pixel data transmitted by the photosensitive element. In the latter case, CogniMem can be used to support other CogniMems and can be used, for example, to identify different types of data transmitted by other CogniMem units (e.g., to incorporate response patterns from multiple cognitive sensors). The following is a list of patents and published applications (the entire contents of each of which are hereby incorporated by reference) for all of the disclosures of the the the the the the the the the the , neuron circuit No. 863; modified neuron circuit architecture No. 5,717,832; circuit 5,701,397 for precharging idle neuron circuits; karaoke circuit for series connection of neuron circuits; Circuit No. 5, 740, 326 for searching/classifying data in neural networks; No. 6, 332, 137 for parallel hardware-related parallel identification; No. 6,606,614 single-line search and classification; Japanese application JP8-171543 for nerves Daisy chain circuit connected by series circuit; JP8-171542 advanced load circuit; JP8-171541 aggregation circuit (search/classification); JP8-171540 neural network and neural chip; JP8-069445 neuron circuit structure; Korean patent application KR164943 innovative neuron circuit architecture; European patent - EP0694852 innovative neuron circuit architecture; EP0694854 modified neural semi-conductor chip architecture EP0694855 neural network search/classification; EP0694853 for pre-charging the input vector component of the idle neuron circuit in the identification phase; EP0694856 for the series connection of the neuron circuit daisy chain circuit; Canada Application CA2 149478 modified neuron circuit architecture; Canadian patent CA2149479 improved neural semiconductor wafer architecture. 112868.doc •10- 200805178 The number of neurons constructed on CogniMem can be changed from 1 to N, where N is theoretically infinite due to the architecture of the neuronal unit. Currently, N can be as high as about 1,000. In general, it is determined by the application, and in particular, n is determined by the diversity of patterns to be recognized and the type of decision to be transmitted. Those skilled in the art will recognize that chopping techniques can be an important factor in determining the number of neurons that can be provided per unit area. 1A and 1B illustrate an exemplary configuration of an image recognition apparatus according to an embodiment of the present invention. Figure 1A is a top plan view of a device 1 comprising a substrate 102 that can be made from a variety of transparent or translucent materials such as glass, plastic glass, transparent plastic, and the like. One or more cognitive sensors 1 〇 4 (in this case, in an array) may be embedded in the substrate 102 or attached or glued to the surface of the substrate 102 as in this case, or other The manner is coupled to the surface of the substrate 102 (see FIG. 1B). An optical path may be etched or deposited on the substrate before each photosensitive element. For example, the lens 102a for each cognitive sensor 〇4 can be generated at the location of the substrate 101's known sensor 104. Alternatively, the microlens 10a can be inserted into the substrate 102 (Fig. 2) or glued to (Fig. 3A - Fig. 3B) the substrate 102 in front of the photosensitive member. Another option may be to change the substrate to change the refractive index of the portion of the substrate adjacent each inductor to focus the incident light. As shown in Fig. 1B, incident light is collected by the substrate lens 102a on each of the cognitive sensors 1〇4. The plurality of lenses 102a allow the cognitive sensor 1 to cover a variety of fields of view, preferably equal to the surface of the substrate, but may also cover % of view that is narrower than or equal to the % of view of the surface of the substrate. The microlens 1 〇 2 a transforms the array of cognitive sensors 1 〇 4 into a telecentric image sensing device with an unrestricted surface and field of view. 112868.doc 11 200805178 Figure 2 is a top plan view of a single stone imaging apparatus in accordance with another embodiment of the present invention. As shown, the lens 1 〇 2a is intended to be inserted into the substrate 1 〇 2 and placed above each of the % sensors 104. As an example of the use of the image forming apparatus, it is shown that the dna piece is placed on the surface of the substrate 1〇2. Each cognitive sensor can be configured to recognize a particular DNA segment individually or in cooperation with a neighboring cognitive sensor 1-4 and output a signal when the segment has been identified. 3A-3B illustrate an exemplary embodiment of an individual cognitive sensor 104. As shown in Fig. 3A, the concentrated neuron region 104a surrounds the pixel sensing region i〇4b. The neurons in neuron region 104a can be coupled to the pixel region [04b in the sensing element] and can be configured to recognize the pattern sensed by pixel region 104b. As shown in FIG. 3B, a convex lens or microlens 1 2a is disposed over the pixel region 104b on the surface of the substrate 1 2 for focusing incident light on the pixel region 1 or directly connected to the inductor There is no intermediate substrate. Lens 丨〇 2 & can be deposited, for example, chemically on the substrate by conventional methods. 4 is a functional block diagram of an exemplary cognitive sensor 1 (10) in accordance with an embodiment of the present invention. The cognitive sensor 101 includes an inductor or sensing area 4, a tribute presentation logic 404, a neural network 406, and an area decision logic 408. Sensor 402 can include one or more sensing elements (such as photosensitive elements). The data presentation logic 404 is coupled to the sensing area 402 and the neural network 406| and is configured to present the data output from the sensor to the neural crest in a manner suitable for processing. Neuron 406 is or becomes taught by knowledge, and can process data input from presentation logic 404 to neuron 4〇6 and output the processed data to regional decision logic 4〇8 based on the processing data generation decision. The region decision logic 408 can be configured by various known methods with other cognitive sensors or c〇gniMem. 112868.doc -12- 200805178. Thus, the cognitive sensor 104 can be configured in an array of arrays and arrays. Figure 5A and Figure 5B show the cognition Configuration of the array of sensors. As shown in FIG. 2, each of the cognitive sensors 104 can be coupled to a plurality of cognitive sensors ι 4 to the array 502. As described below, the input and output bus bars can be used for sensing. For the series or parallel connection, as shown in Figure 5B, each array 5〇2 can be consuming with a plurality of arrays (10) to form an array group 504. #Array of arrays configured with cognitive sensors 1〇4 Second, it produces a highly effective identification device with high resolution and high speed. That is, the resolution of the imaging device can be increased by increasing the number of sensors. However, by providing a stable regional decision-making capability in the form of C〇gniMem, The increase in the number of inductors does not reduce the processing speed of the device. Furthermore, it should be understood that the array can be organized in a number of different geometries, and the invention is not limited to the squares mentioned above, 'per-neurons can be combined with a plurality of input terminals (four) Pick up,

The 2 end 1 η can be, for example, a multi-intelligence input end, but is not limited thereto. Fig. 6A II. The representation of the neurons at the turn-in end is simplified in Fig. 6A. Because a god: use, stream 602 (the bus bar 6.2 does not exist on the map) to group _ each H column ' as shown in the simple parallel architecture of (4). The mother of the neuron—the wheel end can be connected to the global decision bus 4〇6. Figure 7 is a block diagram of the present invention in accordance with the present invention. The functional side of the exemplary neuron of ❸ "" is used to learn and recall the light intensity coded; the nickname is mainly the simplification process of the poor material a / knife cloth. Parallel connections can be made as shown in Figure 6C. Π 2868.doc -13- 200805178 Neurons This means that all neuron inputs and all their outputs are connected in parallel. The data signal can be input to the neuron 700 from the multiplex input bus (not shown). The learning cognitive multiplexer 7〇2 can divide the multiplex input signal and transmit the input data signal to the neuron recall memory 7〇4 and the related logic element 7〇6. The nerve το recalls the memory 7〇4 to process the input signal, and the processed signal

The number is output to the relevant logic element 7〇6. The associated logic element 〇6 includes similar coefficient decision elements 〇6a 〇 each neuron can receive a broadcast pattern (i.e., a vector representing the digital signature of the sensor data) generated by the data presentation logic 〇4. This broadcast pattern can be used to convert data generated by sensors in the transient or in the time domain (data simplification). The 绖 绖 has two possible subsequent timing states: sleep, forthcoming learning (RTL) and subsequent determination. At least—the neurons are in the RTL state at all times unless the network is full (ie, all neurons will be determined). If all parallel connected neurons are considered to be - chains, the RTL neurons can be privately moved from the first position of the key to the last position. In the case indicated here, the rtl neuron will typically be determining the neuron m and the money neuron will be to the right of the RTL neuron. When the neurons are dormant, the complex adenine p y 1 _ , R will not react to any incident pattern. The RTL neurons will be enrolled in the syllabus and loaded into their memory to learn i lL T>r_T ^ during the user process decision learning. This RTL neuron will not participate in the identification process, but will be dedicated to building new knowledge while learning. The learning process involves the creation of new knowledge when an unknown map macro 〇 # ϋ 且 且 且 且 且 且 且 868 868 868 868 868 868 868 868 868 868 868 868 868 868 868 868 868 This knowledge addition will occur when the RTL neuron recognizes the incident pattern (meaning off: indeed, the meta will reduce its similar domain to avoid the class. This situation causes knowledge modification or "adaptive learning." / 曰 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光 光Figure - this pattern experiences the digits that cause the radiation measurement pattern

Data simplification process. The simplification process cannot exceed the so-called "most & To enumerate examples of 5x7 matrices, it is possible to distinguish all European capital letters rather than Chinese characters, which requires a 16χ16 matrix. When it is determined that the neuron is in the RTL state, it associates the vector loaded into the memory memory 704 with the category stored in the class register. When the incident pattern enters the determining neuron, the learning/recognition multi-benefit 702 will transmit it to the associated logic 7〇6 to estimate the similarity of this pattern to the vector stored in the recall memory 704. If the calculated similarity is found to be less than or equal to the similarity factor 7〇6a, then the neuron will be excited, and thus the signal passes through logic 712. The function of the stimulus/suppression logic is to perform global arbitration in all determined "stimulated" when many neurons become excited (meaning that 'encitation ^ neurons are performing global arbitration, and, suppression " does not have the best similarity Each of the neurons in the region of interest can be associated with a region of interest (ROI) for each video frame. Each cognitive sensor can extract the signature of 仏〇1 to broadcast to its neurons. (for learning or identification purposes). R〇][signature is a compressed format of its pixel value 112868.doc -15- 200805178, which is reduced to fit a sequence of N values, where ^ is the neuron memory The size of the unit. An example of the memory capacity of a neuron equipped with 256 bytes. The cognitive sensor can classify the rectangular roj of NxM pixels. The r〇i signature will be compressed from NxM by (for example) simple block compression. The value is reduced to 256. The cognitive sensor can be configured to handle R〇I of any shape, and the signature extraction can be selected for special applications (eg, partial inspection, surface inspection, surface identification, target tracking, etc.) ). Some signatures mention Integration time, repeatability, etc. In addition, the neuron can be equipped with a memory unit larger than 8 bits to supply 12-bit pixel resolution or larger analytic neurons along with the sensor and data to the input of the sensor. The combination of presentation logic constitutes a new method for breaking into image recognition without any software required for the learning or identification process.

The location of the CogniMem can be driven by the response of the pass or selective φ CogniMem unit). (such as by other understanding that the substrate carrying the cognitive sensor acts as a mechanical support and as a lens

Tan or curved surface.

The heart is deducted from a small number of wires. Knowledge loaded into the cognitive sensor: Identification of different series of patterns. It may be better to resolve related or unrelated 112868.doc 16 200805178 Example According to an embodiment of the invention, the cognitive sensor is ideal for performing inspections during the automated manufacturing process. As shown in Figure 8, one or more cognitive sensations can be used to inspect the water bottle. In this example, three different cognitive sensors are used to examine three different regions, called Experts. The global response may be determined by a combination of three "expert" cognitive sensors. In this example, cognitive sensor 1 (expert 1) may be trained to classify the signature of the ROI containing the cap 802. The cognitive sensor 1 may The classification of its size 1 is in two categories: good and bad. The bad category can be combined with several conditions: the cover is missing or the cover is not properly tightened. Similarly, the cognitive sensor 2 (expert 2) can learn with the fluid line in the bottle 8〇4 The signature of the ROI of the father. The ROI can be a narrow vertical rectangle and will ideally cover the minimum and maximum possible fill height in the bottle. Depending on the manufacturer's quality control criteria, the cognitive sensor 2 can classify its R〇I For any number of categories (for example): acceptable and unacceptable; too high, acceptable and too low; or too high, high but acceptable; range, low but acceptable, too low. Cognitive sensor 3 ( The expert 3) can learn the signature of the area of interest that covers the tag area 806. The cognitive sensor 3 can be trained to recognize the diversity of conditions or combinations of conditions, such as, for example, missing tags, defective tags (torture, Scratched or folded), wrong Labels (inverted, tilted) and preferably. The output from the cognitive sensor 1 - 3 can be provided to a controller associated with the automated manufacturing process to take appropriate action based on the decisions made thereby. For example, the cognitive sensor can be individually packaged to form a smart photocell or smart microlens. This device is used in many technologies and can be used, for example, to extract moving parts of the moving part of the mobile transmission, and eight in the mechanical assembly process. In the identification route or in the phase phone I:, points (Fig. 9A); for biometric identification 'such as in the illuminance (Fig. 9C). ' or for the guest detection and identification in the door peephole, etc. The invention provides a driver perception detection system. Referring to FIG. 10, a cognitive sensor 104 can be embedded in a windshield, a vehicle, or a motor vehicle. w 1 plate flat panel display or headlight. Can teach cognitive sensing

Benefits, 1 〇 4 identify the indication drive 0 When the horse is no longer focused (for example, the driver falls asleep) pattern, and the signal is triggered to trigger the alarm. Such patterns may include gaze follow-up, facial recognition, facial expression recognition and more. In addition, it can be taught that the windshield or headlights > Λ e _ Sense sensor 104 recognizes objects or events outside the vehicle 'such as identification of material water m (4), or vehicle accidents for driving accident warning systems . Objects that appear randomly in the far field or near field of view can be detected in the manner of Xu Xi. For example, a, two or three sensors can be equipped with lenses that are focused at different distances. The sensor can carry the same knowledge, but it works for areas of different sizes; If at least the sensor recognizes the object, then the global response of the identification system is considered correct. Also, the cognitive sensor can be designed with input sensors that are sensitive to different wavelengths, such as near-near, IR, filtered color ((3) filtered, etc.). For special features or scenes, these cognitive sensors will produce different pixel values, but can be trained on their individual video images to identify the type of object. In target tracking, the combination of near IR and IR cognitive sensors will Provides the ability to identify goals at any time of the day. 112868.doc -18- 200805178 Another embodiment of the present invention can be used in many other manufacturing applications. For example, as shown in Figure 11A, a one-dimensional array of (10) sensory stealers 1102 can be used to inspect the glass float '(8)(四)fl〇at)U〇3 during the manufacturing process. As shown in FIG. 11B, a two-dimensional array of cognitive sensors, 1104 can be used to make contaminants at the bottom of the container 11〇5 (such as a beverage bottle). In such applications, each cognitive sensor can be taught to identify the indicator glass. A pattern of cracks or fluid contaminants. According to another embodiment of the present invention, a sensor can be configured on a glass plane or the like to perform a plurality of independent functions. Cognitive sensors can be grouped and each group taught with different knowledge. Figure 12 shows an example of a sliding glass door 12〇2 comprising a plurality of cognitive sensor groups (10) for pain sensing different sizes of nearby objects. The first group can be taught by knowledge of the first size 12〇8 of a recognized person or animal (eg, a dog) while the second group can be taught by a different size person (eg, a boy) 121〇, the third group is used for another , size of people (for example, adults) 1212, and so on. Each group of 12〇4 can be coupled to one or more CogniMem 1206 for control of the sliding door. It will be apparent to those skilled in the art that the imaging device of the present invention, as read in this patent document, can be useful in many other applications not listed herein. For example, another application includes permanent damage (four) measurements (texture changes) in dams, bridges, or other man-made structures. The implementation of this application should be apparent from the above description of embodiments of the invention. In addition, power and signal transmission can be wireless (eg, infrared, photocell, induction coil, etc.). Thus, many preferred embodiments have been described above in detail with reference to the drawings. Although the present invention has been described in terms of these preferred embodiments, it is apparent to those skilled in the art that the present invention may be modified and changed in the spirit and scope of the present invention. Alternative construction. BRIEF DESCRIPTION OF THE DRAWINGS [FIG. 1A-FIG. 1 includes a front view and a top view of an array of inductors disposed on glass or plastic glass or other transparent plastic or transparent substrate in accordance with an embodiment of the present invention, in which the glass or 2 is an etched lens in a plastic glass or substrate, FIG. 2 is a top plan view of an array of inductors disposed on a glass or plexus substrate in accordance with an embodiment of the present invention, showing a detected DNA fragment having an etched lens in the glass or substrate; 3A-3B are respectively a side view and a plan view of an inductor mold according to an embodiment of the present invention; FIG. 4 is a block diagram of an inductor according to an embodiment of the present invention; and FIG. 5A is an embodiment of the present invention. Block diagram of the sensor array, FIG. 5B is a block diagram of a sensor array group according to an embodiment of the present invention, and FIGS. 6A-6C illustrate a neural configuration according to an embodiment of the present invention; FIG. 7 is a A block diagram of a neuron of an embodiment of the invention; and 'Fig. 8-12 illustrate an exemplary application of an image recognition apparatus in accordance with an embodiment of the present invention. [Main component symbol description] 100 device substrate 102a lens 112868.doc 200805178 104 cognitive sensor 104a neuron region 104b pixel region - 202 DNA segment - 402 sensor / sensing region 404 data presentation logic 406 neural network 408 region decision logic 5 02 Array 504 Array Group 700 Neuron 702 Learning Cognitive Multiplexer 704 Neuron Recall Memory 706 Related Logic Element 706 a Similarity Coefficient Decision Element • 709 Class Register 712 Logic ^ 802 Cap, 804 Bottle Fluid Line 806 Label Zone 1102 One of the cognitive sensors Array 1103 Glass float 1104 Two-dimensional array of cognitive sensors 1105 Container 112868.doc -21 - 200805178 1202 Sliding glass door 1204 Cognitive sensor group 1206 CogniMem , 1208 First size person or animal 5 1210 People of different sizes 1212 Another size person • 112868.doc -22 -

Claims (1)

200805178 X. Patent Application Park: 1 · An image recognition device comprising: a sensing area embedded in a transparent or translucent substrate or disposed on a transparent or translucent substrate; and a processing element, Coupling with the sensing region, the processing component is embedded in the substrate or disposed on the substrate; the transparent or semi-transparent substrate is formed between the sensing pixels of the incident image to be sensed An optical interface. S long item 1 shirt image recognition device, wherein the sensing area is transparent or translucent. The shirt identification device of claim 1, wherein the substrate comprises glass, plastic glass or other transparent material. An image recognition device for a member 1, wherein the sensing region comprises one or more sensitized 7L pieces, and the processing element comprises one or more cognitive memory elements, wherein: the mother element is configured to be based on a A signal is received at the input, and each of the cognitive memory elements is configured to recognize a pattern by the signals that the photosensitive elements are rotated. 3 The W image recognition device of claim 4, wherein the cognitive memory component is trainable. 6 The image recognition device of the item 4, wherein each of the cognitive memory elements 匕S is coupled to an input side of the input bus and is connected to the _ output by the 出] The plurality of neurons on the end side, God, and the meta-knowledge teach, the knowledge allows the corresponding neuron 112868.doc 200805178 to recognize a signal and perform a decision. 7. (4) An image discriminating device is provided, wherein the substrate comprises a plurality of transmissive, brother P-knife's per-lens portion-to-optical interface providing an inductive pixel or pixel region of the image recognizing device. 8. The image recognition device of π, wherein the lens portion is formed by etching of the substrate. 9. The image recognition device of claim 1, wherein the image recognition device comprises a plurality of the sensing regions of the array structure. 10. An image recognition device comprising: a plurality of cognitive sensors disposed on a transparent or translucent substrate - each sensor comprising: a photosensitive element; and - a training cognitive memory unit, The photosensitive elements are associated; a plurality of optical interfaces formed on the substrate, each of which is coupled to the plurality of cognitive sensors 11 - corresponding cognitive sensing n optics (four). " n. The image recognition device of claim 10, wherein the optical interfaces are lenses formed by (4) the substrate at a position close to each of the cognitive sensors. α is a request-like image recognition device in which each of the cognitive sensors is trainable and configured to recognize various patterns based on incident light. A. The image recognition device of claim H), wherein each of the cognitive memory elements 3 is coupled to an input side of the input input busbar and connected to an output end side by the wheeled busbar surface A plurality of neurons, 112868.doc 200805178 Each of the neurons is taught by a knower, B 4, 5, which allows the corresponding neuron to discriminate a signal and perform a decision. 14. The image of claim 10, 53, wherein the substrate comprises glass, plastic I glass or other transparent material. 15. According to the image recognition of claim 13 - m, #, I, wherein the mother-cognitive memory unit is taught to identify a shadow-shadow unit... different parts' and the plurality of cognitive minds are second, tolerate common Operate to recognize the image. 16. If the image of claim 15 is #% device, where are the cognitive sensors sorrowful? A signal is output during the recognition of the image. $1G shirt image recognition device in which the cognitive sensors operate in parallel at low frequencies. - The image recognition device of claim 10, wherein the cognitive sensors operate at a very low current. An image recognition device, comprising: a substrate; _: a sensing member for sensing various patterns of incident light and rotating a signal based on the sensing patterns, the cognitive sensing member being embedded in the substrate; An optical interface member for providing an optical interface to the cognitive sensing member. 20. The image recognition device of claim 19, wherein the cognitive sensing component comprises a plurality of photodetectors and a plurality of cognitive memory cells, wherein each of the photodetectors is configured to be based at an input The received light is used to rotate a signal, and each of the cognitive memory units is configured to recognize a pattern according to the special signal that is emitted by the optical detectors at 112868.doc 200805178. 21. The image recognition, fly set of claim 19, wherein the optical interface member comprises a plurality of lenses formed in the substrate. The image recognition device of claim 19, wherein the cognitive sensing component comprises: a plurality of cognitive sensors, each of which comprises a photosensitive element; and a training § 忍 知 恃 _ _ C > Corresponding to the photosensitive element, the image recognition device of claim 22, wherein each of the cognitive memory cells comprises a multiplexed input busbar connected to its input side and outputted by an output The busbar is coupled to one of the plurality of neurons on the side, said side, each of which is known, for teaching, the knowledge allowing the corresponding neuron to recognize a signal and perform a decision. For example, the image recognition device of the present invention, wherein the cognitive sensing component performs image recognition operation through a plurality of digitization of the row components without a software program, the self-sufficient behavior of self-sufficiency. The image recognition device of claim 1, wherein the image recognition device is configured to output a wireless output signal. 26. The image recognition device of claim 1G, wherein each "Cognitive Sensor" is configured to transmit and receive a plurality of wireless signals. 27. The image recognition device of 4 (4), wherein the cognitive sensing component is configured to transmit and receive a plurality of wireless signals. 28. The image recognition device of the present invention, wherein the device is wirelessly powered by 112868.doc 200805178. 29. The image recognition device of claim 1 wherein the device is wirelessly powered. 3. The image recognition device of claim 19 is 0' wherein the device is wirelessly powered. 31. An image recognition device is presented A combination of logic. Contains neurons with a sensor and resources. The method includes the following steps: forming an image recognition method on a substrate or depositing on a substrate, embedding the cognitive sensing component in the cognitive sensing component for sensing a pattern of light emitted by a plurality of people, and outputting a signal based on the sensing patterns; An optical interface member for providing an optical interface to the cognitive sensing member. 33. The image recognition method of claim 32, wherein the cognitive sensing component comprises a plurality of photodetectors and a plurality of cognitive memory cells, Each of the photodetectors is configured to output a signal based on light received at an input, and each of the cognitive memory cells is configured to output the signals from the photodetectors Identify a pattern in the middle. 34. The image recognition method of claim 32, further comprising the step of providing a plurality of lenses formed in or deposited on the substrate as part of the optical interface member. 35. The image recognition method of claim 32, wherein the cognitive sensing component comprises: a plurality of cognitive sensors, each comprising: 112868.doc 200805178 coupling photosensitive element; and a trainable cognitive memory unit, and the photosensitive element Corresponding to each of the cognitive memory singles 36. The image recognition method of claim 35, wherein the element comprises a multiplex input sink a, which is connected to one of the input sides and is connected to the output block by an output bus A plurality of neurons on the creep side are extracted, each of which is subject to a knowledge I, which allows the corresponding neuron to recognize a signal and perform a decision. 112868.doc