AU685904B2 - Produce recognition system - Google PatentsProduce recognition system Download PDF
- Publication number
- AU685904B2 AU685904B2 AU15034/95A AU1503495A AU685904B2 AU 685904 B2 AU685904 B2 AU 685904B2 AU 15034/95 A AU15034/95 A AU 15034/95A AU 1503495 A AU1503495 A AU 1503495A AU 685904 B2 AU685904 B2 AU 685904B2
- Prior art keywords
- 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.)
- Expired - Fee Related
- 238000005286 illumination Methods 0.000 claims description 30
- 244000144725 Amygdalus communis Species 0.000 description 11
- 235000020224 almond Nutrition 0.000 description 11
- 240000007119 Malus pumila Species 0.000 description 8
- 238000000034 methods Methods 0.000 description 8
- 235000011437 Amygdalus communis Nutrition 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 240000002840 Allium cepa Species 0.000 description 2
- 240000008067 Cucumis sativus Species 0.000 description 2
- 240000002254 Ananas comosus Species 0.000 description 1
- 235000007119 Ananas comosus Nutrition 0.000 description 1
- 240000005781 Arachis hypogaea Species 0.000 description 1
- 235000010777 Arachis hypogaea Nutrition 0.000 description 1
- 235000017647 Brassica oleracea var italica Nutrition 0.000 description 1
- 241000207199 Citrus Species 0.000 description 1
- 240000000560 Citrus x paradisi Species 0.000 description 1
- 240000008966 Cynara cardunculus Species 0.000 description 1
- 240000002860 Daucus carota Species 0.000 description 1
- 235000002767 Daucus carota Nutrition 0.000 description 1
- 240000009164 Petroselinum crispum Species 0.000 description 1
- 235000016761 Piper aduncum Nutrition 0.000 description 1
- 235000017804 Piper guineense Nutrition 0.000 description 1
- 240000000129 Piper nigrum Species 0.000 description 1
- 235000008184 Piper nigrum Nutrition 0.000 description 1
- 241000758706 Piperaceae Species 0.000 description 1
- 240000001016 Solanum tuberosum Species 0.000 description 1
- 235000002595 Solanum tuberosum Nutrition 0.000 description 1
- 240000007329 Zingiber officinale Species 0.000 description 1
- 235000006886 Zingiber officinale Nutrition 0.000 description 1
- 230000001276 controlling effects Effects 0.000 description 1
- 239000000203 mixtures Substances 0.000 description 1
- 235000014571 nuts Nutrition 0.000 description 1
- G06—COMPUTING; CALCULATING; COUNTING
- G06K—RECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
- G06K9/00—Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
- G06K9/36—Image preprocessing, i.e. processing the image information without deciding about the identity of the image
- G06K9/46—Extraction of features or characteristics of the image
- G06K9/4642—Extraction of features or characteristics of the image by performing operations within image blocks or by using histograms
- G06K9/4647—Extraction of features or characteristics of the image by performing operations within image blocks or by using histograms summing image-intensity values; Projection and histogram analysis
- G07G—REGISTERING THE RECEIPT OF CASH, VALUABLES, OR TOKENS
- G07G1/00—Cash registers
- G07G1/0036—Checkout procedures
- G07G1/0045—Checkout procedures with a code reader for reading of an identifying code of the article to be registered, e.g. barcode reader or radio-frequency identity [RFID] reader
- G07G1/0054—Checkout procedures with a code reader for reading of an identifying code of the article to be registered, e.g. barcode reader or radio-frequency identity [RFID] reader with control of supplementary check-parameters, e.g. weight or number of articles
- G06—COMPUTING; CALCULATING; COUNTING
- G06K—RECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
- G06K2209/00—Indexing scheme relating to methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
- G06K2209/17—Recognition of food, fruit, vegetables
IP I I 1~ Ib Y0994056 PRODUCE RECOGNITION SYSTEM FIELD OF THE INVENTION This invention relates to the field of recognizing identifying, classifying, grading, and verifying) objects using computerized octical scanning devices. More specifically, the invention is a trainable system and method relating to recognizing bulk items using image processing.
BCKGROUND OF THE INVENTION ";*ooo Image~ processing systems exist in the prior art for recognizing objects. Often these systems use histograms to per- "IS form this recognition. One common histogram method either develops a gray scale histogram or a color histogram from a .(color) image containing an object. These histograms are then compared directly to histograms of reference images.
Alternatively, features of the histograms are extracted and compared to features extracted from histograms of images containing reference objects.
st I~L~ aI r alr II an- I Y0994056 The reference histograms or features of these histograms are typically stored in computer memory. 'The prior art often performs these methods to verify that the target object in image is indeed the object that is expected, and, possibly, to grade/classify the object according to the quality of its appearance relative to the reference histogram. An alternative purpose could be to identify the target object by comparing the target image object histogram to the histograms of a number of reference images of objects.
.0 In this description, identifying is defined as determining, given a set of reference objects or classes, which reference object the target object is or which reference class the .9 target object belongs to. Classifying or grading is defined 0 as determining that the target object is known to be a certain object and/or that the quality of the object is some quantitatively value. Here, one of the classes can be a "reject" class, meaning that either the quality of the object is too poor, or the object is not a member of the known class. Verifying, on the other hand, is defined as determining that the target is known to be a certain object or class and simply verifying this to be true or false. Recog- L~1 W I a 1 I 111111 -sl 11 1111119 11 Y0994056 nizing is defined as identifying, classifying, grading, and/or verifying.
Bulk items include any item that is sold in bulk in supermarkets, grocery stores, retail stores or hardware stores.
Examples include produce (fruits and vegetables), sugar, coffee beans, candy, nails, nuts, bolts, general hardware, parts, and package goods.
In image processing, a digital image is an analog image from a camera that is converted to a discrete representation by dividing the picture into a fixed number of locations called picture elements and quantizing the value of the image at those picture elements into a fixed number of values. The resulting digital image can be processed by a computer algorithm to develop other images. These images can be stored in memory and/or used to determine information about the imaged object. A pixel is a picture element of a digital image.
Image processing and computer vision is the processing by a computer of a digital image to modify the image or to obtain ,ql IIC1 Y0994056 from the image properties of the imaged objects such as ob- ject identity, location, etc.
An scene contains one or more objects that are of interest and the surroundings which also get imaged along with the objects. These surroundings are called the background. The background is usually further away from the camera than the object(s) of interest.
Segmenting (also called figure/ground separation) is separating a scene image into separate object and background images. Segmenting refers to identifying those image pixels that are contained in the image of the object versus those that belong to the image of the background. The segmented object image is then the collection of pixels that comprises 'the object in the original image of the complete scene. The ;*area of a segmented object image is the number of pixels in the object image.
Illumination is the light that illuminates the scene and ob- jects in it. Illumination of the whole scene directly determines the illumination of individual objects in the scene I' -II plll~Ty~e~ Sy0994056 and therefore the reflected light of the objects recei,Ved by imaging apparatus such as video camera.
Ambient illumination is illumination from any light source except the special lights used specifically for imaging an object. For example, ambient illumination is the illumination due to light sources occurring in the environment such as the sun outdoors and room lights indoors.
Glare or specular reflection is the high amount of light ree.
flected off a shiny (specular, exhibiting mirror-like, pos- .I sibly locally, properties) object. The color of the glare is mostly that of the illuminating light (as opposed to the natural color of the object).
A feature of an image is defined as any property of the image, which can be computationally extracted. Features typi- Scally have numerical values that can lie in a certain range, say, RO Rl. In prior art, histograms are computed over a whole image or windows (sub-images) in an image. A histogram of a feature of an image is a numerical representation of the distribution of feature values over the image s I IIL a ICI C r I- .I_ Y0994056 or winu A histogram of a feature is developed by dividing the feature range, RO Rl, into M intervals (bins) and computing the feature for each image pixel. Simply counting how many image or window pixels fall in each bin gives the feature histogram.
Image features include, but are not limited to, color and texture. Color is a two-dimensional property, for example Hue and Saturation or other color descriptions (explained SLO below) of a pixel, but often disguised as a three- Sdimensional property, the amount of Red, Green, and Blue (RGB). Various color descriptions are used in the prior art, including: the RGB space; the opponent color space; the Munsell color space; and, (4) the Hue, Saturation, and Intensity S,I) space. For the S. <o
latter, similar to the Munsell space, Hue refers to the color of the pixel (from red, to green, to blue), Saturation is the "deepness" of the color from greenish to deep saturated green), and Intensity is the brightness, or what the pixel would look like in a gray scale image.
Texture, on the other hand, is an visual image feature that 7 i I
Y0994056 is much more difficult to capture computationally and is a i feature that cannot be attributed to a single pixel but is attributed to a patch of image data. The texture of an image patch is a description of the spatial brightness variation in that patch. This can be a repetitive pat tern (of texels), as the pattern on an artichoke or pineapple, or, can be more random, like the pattern of the leaves of parsley. These are called structural textures and statistical textures, respectively. There exists a wide range of textures, ranging from the purely deterministic arrangement of a texel on some tesselation of the two-dimensional plane,: fe e to "salt and pepper" white noise. Research on image texture has been going on for over thirty years, and computational measures have been developed that are one-dimensional or i higher-dimensional. However, in prior art, histograms of texture features are not known to the inventors.
Shape of some boundary in a5n image is a feature of multiple boundary pixels. Boundary shape refers to local features, such as, curvature. An apple will have a roughly constant curvature boundary, while a cucumber has a piece of low curvature, a piece of low negative curvature, and two pieces of I L~ILb Y0994056 high curvature (the end points). Other boundary shape measures can be used.
Some prior art uses color histograms to identify objects.
Given an color image of the target object, the color representation used for the histograms are the opponent color: rg R G, by 2 B R G, and wb R G B.
The wb axis is divided into 8 sections, while rg and by axes 4 are divided into 16 sections. This results in a threedimensional histogram of 2048 bins. This system matches target image histograms to 66 pre-stored reference image histograms. The set of 66 pre-stored reference image S: histogram is fixed, and therefore it is not a trainable system, unrecognized target images in one instance will not be recognized in a later instance.
U.S. Patent 5,060,290 to Kelly and Klein discloses the grading of almonds based on gray scale histograms. Falling almonds are furnished with uniform light and pass by a lin- -8- -sl C -L -a-ri I- Y0994056 ear camera. A gray scale histogram, quantized into 16 levels, of the image of the almond is developed. The histogram is normalized by dividing all bin counts by 1700, where 1700 pixels is the size of the largest almond expected. Five features are extracted from this histogram: gray value of the peak; range of the histogram; number of pixel, :t peak; number of pixels in bin to the right of peak; and, rnamber of pixels in bin 4. Through lookup tables, an eight digit code is developed and if this code is ;L0 in a library, the almond is accepted. The system is not trainable. The appearances of almonds of acceptable quality are hard-coded in the algorithm and the system cannot be trained to grade almonds differently by showing new instances of almonds.
9 U.S. Patent 4,735,323 to Okada et al. discloses a mechanism for aligning and transporting an object to be inspected.
The system more speciftcally relates to grading of oranges.
The transported oranges are illuminated with a light within a predetermined wavelength range. The light reflected is received and converted into an electronic signal. A level histogram divided into 64 bins is developed, where IL-~1La~ 9 I Y0994056 Level (the intensity of totally reflected light) (the intensity of green light reflected by an orange) The median, N, of this histogram is determined and is considered as representing the color of an orange. Based on N, the orange coloring can be classified into four grades of "excellent," "good," "fair" and "poor," or can be graded finer. The system is not trainable, in that the appearance of the different grades of oranges is hard-coded into the algorithms. The use of gray scale and color histograms is a very effective method for grading or verifying objects in an image.
105' The main reason for this is that a histogram is very compact representation of a referenc& object that does not depend on the location or orientation of the object in the image.
However, for image histogram-based recognition to work, certain conditions have to be satisfied. It is required I'iat: the size of the object in the image is roughly known, there is relatively little occlusion of the object I ~~U Y05)94056 most of the object is in the image and not obscured by other objects), ther:e is little difference in illumination of the scene of which the images (reference and target images) are taken from, which the re ference object histo grams and target object histograms are developed,an the obj e ct can be ea-sily segmented out from the background or there is relatively little distraction in the background. Under these conditions, comparing a target object image histogram with reference object image histograms has been achieved in numerous ways in the prior art.
STATEMENT OF PROBLEMS WITH THE PRIOR ART Soepiratmacigsses.n.ehdcli ob o ~i3some prorhs art machngsytesind mento, caim nt bcobstold to e disratons fin then akgoud valriafton invination for obtaining the reference object histograms is different from the color of the illumination when obtaining the target object image histogram. The RGB values of an im- -11- .YO094056 age point in an image are very dependent on the color of the illumination (even though humans have little difficulty naming the color given the whole image). Consequently the color histogram of an image can change dramatically when the color of the illumination (light frequency distribution) changes. Furthermore, in these prior art systems the objects are not segmented from the background, and, therefore, the histograms of the images are not area normalized. This means the objects in target images have to be the same size c 3 the objects in the reference images for accurate recogni- S* tion because variations of the object size with respect to the pixel size can significantly change the color histogram.
It also means that the parts of the image that correspond to the background have to be achromatic black), or ,at least, of a coloring not present in the object, or they will significanly perturb the derived image color histogram.
'3 9. Prior art such as that disclosed in U.S. Patent 5,060,290 fail if the size of the almonds in the image is drastically different than expected. Again, this is because the system does not explicitly separate the object from its background.
This system is used only for grading almonds: it can not -12i a: I Ii)- I I I I Y0994056 distinguish an almond from (say) a peanut.
Similarly, prior art such as that disclosed in U. S. Patent 4,735,323 only recognizes different grades of oranges. A reddish grapefruit might very well be deemed a very large 7 orange. The system is not designed to operate with more than one class of fruit at a time and thus can make do with weak features such as the ratio of green to white reflectivity.
In summary, much of the prior art in the agricultural arena, typified by U.S. patents 4,735,323 and 5,060,290, is concerned with classifying/grading produce items. This prior art can only classify/identify objects/products/produce if they pass a scanner one object at a time. It is also required that the range of sizes (from smallest to largest possible object size) of the object/product/produce be known beforehand. These systems will fail if more than one item is scanned at the same time, or to be more precise, if more than one object appears at a scanning position at the same time.
1 ^*t -13- -I I Y0994056 Further, the prior art often requiires carefully engineered and expensive mechanical environment with carefully controlled lighting conditions where the items are transported i to predefined spatial locations. These apparatuses are designed specifically for one type of shaped object (round, oval, etc. and are impossible or, at least, not easily modified to deal with other object types. The shape of the objects inspires the means of object transportation and it is impossible or difficult for the transport means to transport different object types. This is especially true for oddly ii shaped objects like broccoli or ginger. This, and the use of features that are specifically selected for the particular objects, does not allow for the prior art to distinguish between types of produce.
Additionally, none of the prior art are trainable systems where, through human or computer intervention, new items are learned or old items discarded. That is, the systems can not be taught to recognize objects that were not originally programmed in the system or to stop recognizing objects that were originally programmed in the system.
1 1 1 1 *f Y0994056 One area where the prior art has failed to be effective is -ii in produce check out. The current means and methods for checking out produce poses problems. Affixing (PLU-price lookup) labels to fresh produce is disliked by customers and produce retailers/wholealers. Pre-packaged produce items i are disliked, because of increased cost of packaging, disposal (solid waste), and inability to inspect produce quality in pre-packaged form.
The process of produce check-out has not changed much since the first appearance of grocery stores. At the point of sale (POS), the cashier has to recognize the produce item, weigh or count the item(s), and determine the price. Currently, in most stores the latter is achieved by manually entering the non-mnemonic PLU code that is associated with the produce. These codes are available at the POS in the form of printed list or in a booklet with pictures.
1 e Multiple problems arise from this process of produce check- 0 out: Losses incurred by the store (shrinkage). First; a L Y0994056 cashier may inadvertently enter the wrong code number, If this is to the advantage of the customer, the customer will be less motivated to bring this to the attention of the cashier. Second, for friends and relatives, the cashier may purposely enter the code of a lower-priced produce item (sweethearting).
Produce check-out tends to slow down the check-out process because of produce identification problems.
Every new cashier has to be trained on produce names, produce appearances, and PLU codes.
O* o OBJECTS OF THE INVENTIO: 15 g 0 An object of this invention is an improved apparatus and method for recognizing objects such as produce.
S An object of this invention is an improved trainable apparatus and method for recognizing objects such as produce.
-16i i i P- II Y0994056 Another object of this invention is an improved apparatus and method for recognizing and pricing objects such as produce at the point of sale or in the produce department.
A further object of this invention is an improved means and method of user interface for automated product icentification, such as, produce.
SUMMARY OF THE INVENTION The present invention is a system and apparatus that uses image processing to recognize objects within a scene. The system includes an illumination source for illuminating the 9 scene. By controlling the illumination source, an image processing system can take a first digitized image of the scene with the object illuminated at a higher level and a second digitized image with the object illuminated at a lower level. Using an algorithm, the object(s) image is novelly segmented from a background image of the scene by a comparison of the two digitized images taken. A processed image (that can be used to characterize features) of'the -17- IY-- I ~d~LL~ Id a Y0994056 object(s) is then compared to stored reference images. The object is recognized when a match occurs.
Processed images of an unrecognized object can be labeled with identity of object and stored in memory, based on certain criteria, so that the unrecognized object will be recognize when it is imaged in the future. In this novel way, the invention is taught to recognize previously unknown objects.
Recognition of the object is independent of the size or number of the objects because the object image is novelly normalized before it is compared to the reference images.
C C 5 Optionally, user interfaces and apparatus that determines
other features of the object (like weight) can be used with the system.
BRIEF DESCRIPTION OF THE DRAWINGS e Figure 1 is a block diagram of the one preferred embodiment of the present system.
-18i ii s- Y0994056 Figure 2 is a flow chart showing on preferred embodiment of the present method for recognizing objects.
Figure 3 illustrates segmenting a scene into an object image and a background image.
Figure 4 is a block diagram of a preferred embodiment of apparatus for segmenting images and recognizing object in images.
Figure 5 is afl-low chart-of a preferred method for segmenting target object images.
e we 4 Figure 6 is a flow chart showing a preferred method of char- 5 acterizing reference ot target object feature(s).
Figure 7 is a flow chart showing a preferred method for (area/length) normalization of object feature(s) characterization.
0 Figure 8 illustrates the comparison of an area/length normalized target object characterization to one or more area -19- I I eL i Y0994056 1 normalized reference object characterizations.
Figure 9 is a flow chart showing a preferred (algorithmic) method of training the present apparatus to recognize new images.
Figure 10 is a block diagram showing multiple features of an object being extracted.
Figure 11 i. a flow chart showing the histogramming and nor- I- malizing-of the feature of texture.
Figure 12 is a flow chart showing the histogramming and nor- Ot malizing of the feature of boundary shape.
Figure 13 is block diagram showing a weighing device.
r* Figure 14 shows an image where the segmented object has two distinct regions determined by segmenting the obj3ect image 26 and where these regions are incorporated in recognition algorithms.
Figure 15 shows a human interface to the present apparatus which presents an ordered ranking of the most likely identities of the produce being imaged. L Figure 16 shows a means for human determination of the identity of object(s) by browsing through subset(s) of all the previously installed stored icon images, and the means by which the subsets are selected.
Figure 17 is a preferred embodimeit of the present invention using object weight to price object(s).
DETAILED DESCRIPTION OF THE INVENTION 1 The apparatus 100 shown in Figure 1 is one preferred embodiment of the present invention that uses image processing to automatically recognize one or more objects 131.
A light source 110 with a light frequency distribution that .2 is constant over time illuminates the object 131. The light is non-monochromatic and may include infra-red or ultra violet frequencies. Light being non-monochromatic and of a -21e I- ill, Y6994056 constant frequency distribution ensures that the color appearance of the objects 131 does not change due to light variations between different images taken and that stored images of a given object can be matched to images taken of that object at a later time. The p.eferred lights are flash tubes Mouser U-4425, or two GE cool-white fluorescent bulbs (22 Watts and 30 Watts), GE FC8T9-CW and CE FC12T9-CW, respectively. Such light sources are well known.
A video input device 120 is used to convert the reflected light rays into an image. Typically this image is two dimensional. A preferred video input device is a color camera but any device that converts light rays into an image can be used. These cameras would include CCD camera and CID cameras. The color camera output can be RGB, HSI, YC, or any other representation of color. One preferred camera is a 4 Sony card-camera CCB-C35YC or Sony XC-999. Video input devices like this 120 are well known.
Color images are the preferred sensory modality in this invention. However, other sensor modalities are possible, intra-red and ultra-violet images, smell/odor (measur- -22- -I c~ I Y0994056 able, with mass spectrometer), thermal decay properties, ultra-sound and magnetic resonance images, DNA, fundamental frequency, stiffness and hardness. These modalities can be enabled With known methods of illuminating, measuring, or taking samples of the object 131 and with a compatible imaging device 120 for creating the image.
The object 131 is the object being imaged and recognized by the system 100. The object 131 can comprise one or more items. Although it is preferred that objects 131 be of one type (variety), one or more apples, the items can be of different types, a cereal box (Object A) and an apple (Object System 100 will then recognize objects as !either as Object A, Object B, both Object A and e.
Object B, or, reject objects as unrecognizable. The object(s) can be virtually anything that can be imaged by the system 100, however preferred objects 131 are bulk items including produce (fruits and vegetables), hardware, boxed goods, etc.
A calculating device 140, typically a computer 140, is used to process the image generated by the video input device 120 -23- I -r I I= I I 1 1 Y0994056 and digitized (to be compatible with the computer 140) by a frame grabber 142.
The processing is performed by an algorithm 200. Other calculating devices 140 include: personal computers, and work- stations. The calculating device 140 can also be one or more digital signal processors, either stand-alone or installed in a computer. It can also be any special hardware capable of implementing the algorithm 200. A preferred embodiment is a Datatranslation DSP board DT 2878 coupled to a Datatranslation DT 2871 frame grabber board residing in an IBM ValuePoint computer, or in the IBM 4690 series of POS Cash Recisters. The frame grabber 142 is a device that O* eo digitizes the image signal from the camera 120. If the cam- 1. era 120 is a digital camera then a separate fraam grabber 142 may not be required. The digitizer may be separate from the computer 140 or an integrated part of it. The image may be stored in a standard memory device 144. Given this disclosure, one skilled in the art could develop other equiv- 9. alent calculating devices 140 and frame grabbers 142.
An optional interactive output device 160 can be connected -24- L I IL I I L L L du I Y0994056 to the calculating device 140 for interfacing with a user, like a cashier. The output device 160 can include screens that assist the user in decision making 164 and can also provide mechanisms to train 162 system 100 to recognize new objects. An optional weighing device 170 can also provide an input to the calculating device 140 about the weight (or density) of the object 131. See description below (Figure 13).
il-[ i a a: a *r a.
*r 54 0 a0 S Si a Figure 2 is a flow chart of the algorithm 200 run by the calculating device, or computer 140. In step 210, a target object to be recognized is imaged by camera 120. Imaging like this is well known. The. image of target object 131 is then novelly segmented 220 from its background. The purpose of step 220 is to separate the target object 131 from the background so that the system 100 can compute characteristics of separated object 131 image pixels independently of the background of the scene. In step 230 one or more features of the object 131 can be computed, preferably pixel by pixel, irom the segmented object image. In step 240, characterizations of these pixel-by-pixel computed feature sets are developed. Normalizing, :in step 250, ensures that these
:13 ir: i i'
1 Y0 994056 characterizations do not depend on the actual area, length, size, or characteristics related to area/length/size that the object(s) 131 occupy in the image, so that one or multiple instances of object 131 are recognized as same object type. Preferred means of normalizing the characterization by the segments occupied by objects 131 in the image is achieved by counting the number of times feature characteristic are computed. (This is described further in Figure 7. The preferred means of normalizing is by area or by length.) In step 260 the count-normalized characterization of the target object is compared with the count- normalized characterizations of reference objects, which are stored in memory storage 270. The storage 270 may be lo- 15 cated in the storage device 144 or computer 140. (See the S description in Figure In step 251 area-normalized characterizations are stored, depending on certain criteria 255 in computer'memory 270. This step enables the system 100 to be trained, since the storage criteria 255 might permit storage 251 of new reference images which can later be compared to target 131 images. (See the description of Figure -26i Step 220 is the segmenting or separating of the object image from the background image. This step is performed so that the features of the target object can be processed independ- i ently of the effects and disturbances of the background Iof the scene. Figure 3 illustrates two preferred methods (Fig- ure 3a and Figure 3b, respectively) that segment the object image from the background image.
Figure 3a shows two scenes. The first imaged scene 310, shows an image of a background 311 without any other objects present in the field of view of camera 120. The second imaged scene 320 includes both an image of the scene background 311 and an image 130 of one or more objects 131.
Here the pixels of the imaged object 130 replace pixels in the background image 311 in those areas of the scene image 320 where object 131 is present. Hence, it is an image of background 311 with instances of objects 131 present in the scene.:-: A comparison of the scenes 310 and 320, preferably on a pixel by pixel basis, allows the object image 130 to be segmented (separated out) from the background image 311 of the -27- I ~I i m -rr rr Y0994056 scene. If for a given pixel in the 320 image, the brightness is different from more then) the image brightness of same pixel in 310, this pixel belongs to object image 130. If for a given pixel in the image 320, the brightness is equal to same pixel in 310, this pixel belongs to background image 311. (See the description of Figure Figure 3b shows two images of a scene with a background and one or more objects produced by a preferred embodiment of this invention that enables segmentation of the object image. Image 330 in Figure 3b is an image of a scene (includ- ing objects 131 and a background 311) with light source 110 off. That is, the scene image 330 consist of an image of background 311 illuminated by ambient light. Also in the 1 scene image 330 are the object images 135 obscuring the background. Because the light source 110 is off, object images 135 appear dark in scene image 330 because they are not illuminated by the light source 110.
Image 340 in Figure 3b is an image of the scene with light: source 110 on. In this case, the light source 110 illumi- -28- I I iX.i- Y0994056 nates objects 131 in field of view of camera with an amount of light greater than ambient light. This results in the object images 130 being brighter (than in 330) in scene image 340.
Figure 4 is a block diagram showing a preferred system 400 for imaging scenes (such as those described in Figure 3), segmenting object images 130 from their background image 311 of the physical background 312, and recognizing object(s) 131. A general description of component 401 of this system 400 can be found in U. S. Patent Application number xxx, entitled "Object Imaging System" filed in April, 1994 which is i herein incorporated by reference.
I The preferred system 400 places the object 131 above light 110 and camera 120, thus providing images of object 131 looking up from below. The system 400 provides a support 405 for the object 131 and also ensures that object is of fixed and repeatable distance 407 from camera 120. In addition, the system 400 allows imaging of shiny object (like a plastic bag) with reduced glare (specular reflections) in the image by providing a filtering system comprised of 410 -29i 1. -3
Y0994056 and 420.
The system 400 comprises an opaque enclosure 401 for the light 110 and camera 120. The enclosure has a single opening 403 facing the object 131. The opening 403 is of a sufficient size to allow the object 131 to be imaged by the camera 120 and illuminated by the light 110. The opening can be square, round or any other shape. A transparent surface 405 covers the opening 403 in the enclosure 401. This surface 405 could be a sheet of' glass. The transparent surface 405 provides a support on which the imaged object 131 can be placed. By placing the object 131 on the surface: i 405, the distance 407 between camera 120 and object 131 ree* mains fixed thus providing the means for repeatable imaging.
To remove glare from image of object 131 (from object 131 itself and possibly a surrounding plastic bag) a polarizing filter 420 is incorporated with the lens of camera 120 or placed just above the lens of the camera 120. A second .i1:1 polarizing filter 410 is placed between the light 110 and the opening 403. This insures that the light reaching the object 131 is polarized. Alternatively, the light may be
II-- -I_ Y0994056 completely enclosed by the polarizer. If the light is partly enclosed in a box (such as a camera flash) or by a reflector (such as a photographic spotlight) the polarizer needs to be placed only on the opening in the light assembly which allows the light through. The direction of the polarization in first filter 410 is orthogonal to the direc- tion of polarization in second filter 420. It is well-known from prior art that specular reflection reflected off an object (such as object 131) is polarized as opposed to the diffuse (matte) reflection reflected off the object. Imaging object 131 with a polarizing filter 420 thus reduces glare in image. Further, illuminating 131 with light 116 Ue ot polarized by 410 reduces the amount of glare on object 131.
1: 410 also ensures that the polarization angle of the reflected specular light, off object 131 is orthogonal to polarizer 420. Hence, imaging object 131 through polarizer 420 which is orthogonal to polarizer 410 further reduces the amount of glare in object image 130.
S e A light Control 450 switches the light 110 on and off, or switches light 110 between different intensity levels. The i I I Y0994056 control 450 may be implemented on the computer 110 or be I connected directly to the light 110 or may be a separate device. The control may be a part of the light 110 as a timing device such ak. in a strobe. The control may be synchronized with the camera or the computer or both. Light switching controls 450 are well known.
The segmenting step 220 of Figure 2 is further described in Figure 5, which shows a preferred method for segmenting the object image from the scene.
In step 510, an image (a first image) of the scene is produced with the light 110 switched on, or at a higher intensity, so as to illuminate object 131 properly. Control 450 controls the light 110 switching.
In step 520, a second image of the scene is produced with the light 110 switched off or set to a level below the level in step 510. The setting of the light 110 should be such that the object 131 appears darker in the second image than in the first image. By performing these novel steps, the object image 130 can be separated or segmented from the -32- Y0994056 background image 311 in the steps below.
Further, the object 131, the background 312, and the image input device 120 should be at the same position in both step 510 and 520 to assure that the first and second images are in spatial registration. Suppose each pixel is numbered starting in the upper left corner of the image then proceeding across the first line then down to the second line in the manner of reading a book. Registration means that each numbered pixel in the first image corresponds to the same area of the scene (object(s) 131 and background 312) as the identically numbered pixel in the second image. Proper registration can be ensured by either acquiring the first and 6 second image in quick succession, or by imaging a stationary object 131 against a stationary background 312.
The order of acquiring the first and second image may be rea a a versed; that is, step 520 can be performed before step 510.
.26 4 In step 530 of the algorithm 220, the first and second images are digitized in the frame grabber 142. In the com- -33- SY0994056 puter 140, each and every pixel in the first digitized image is compared to the respective pixel at the same location in the second digitized image. Pixel by pixel comparisons such as this are known in the image processing art. For example, although the pixels in each pair being compared must correspond to one another be in the same respective location in each image), the corresponding pixel pairs in the images can be compared in any order. Further, alternative comparisons can be made, comparing ever second or third pixel.
In step 540, a check is performed on a pixel by pixel basis
to determine if a pixel in the first image is brighter than the corresponding pixel in the second image by more than a a. value T. In any pixel comparison, if the pixel in the first image pixel is brighter than its corresponding pixel in the second image by more than T, the algorithm 220 takes the branch 542 and designates this pixel as corresponding to the object 131. Likewise, if the pixel comparison shows that the pixel in the first image is not brighter than its corresponding pixel in the second image by more than the value T, the algorithm 220 takes the branch 544 and designates this -34- I 4 II II Y0994056 pixel as corresponding to the image 311 of physical back- ground 312.
The value of tolerance T may be a constant. A preferred tolerance T is 5% of the largest image intensity. Alternatively, the value of T may vary depending on the positions of pixels in the image or depending on the intensity of the pixel in the dark image. The positional variation of T allows the system to compensate for uneven illumination from source 110. The dark intensity variation of T allows the system to correctly identify foreground objects with low reflectivities (such as black objects). The value T may be fixed or may be recomputed from time to time by the system.
It might, for instance, be necessary to change the value of 6 T as light source 110 ages or changes in intensity for some other reason (such as a variation in the AC line voltage supplied to the bulb). This recomputation could be performed on a pair of images of the background with no object (one image of the background 312 highly illuminated and one 'less so). Since no object is present, both background images should appear to be illuminated the same amount (with ambient light). However, in practice, the light 110 might Y099405 6..
illuminate the background 312 slightly when the light is switched to a higher intensity. Therefore a tolerance T is chosen for the comparison of the corresponding pixel pairs.
The tolerance T could then be set so that only a very small number of pixels in this pair of background images actually passes the test. For example, in a preferred embodiment, T would be set so that fewer than 10% of the pixel pairs differ in illumination more than the tolerance T.
In a preferred embodiment, the steps 530 and 540 are performed on a pixel by pixel basis for each pixel location in the scene image. The result is that the pixels corresponding to the object 131 are collected in a segmented object image 130. Specifically, in the segmented object image, all pixels from the first image that are substantially brighter than their corresponding pixel in the second image are collected in segmented object :;Lmage at the position they were in the first image. Therefore, the segmented object image a *2G corresponds to the desired image of the object 131 removed from the background 312. If needed, the remaining pixels in the image the pixels not corresponding to the object -36- Y0994056b 130) can be assigned any desired value and/or ca'.i be further processed using known image processing techniques.
In like manner, the pixels corresponding the background 312 are collected in a segmented background image 311. Specifically, all pixels from the first image that are not substantially brighter than the corresponding pixel in the second image are collected in the segmented background image at the position they Were in the first image. (In a preferred embodiment, "substantially brighter" means that the
difference in illumination between the pixels in the corresponding pixel pair is greater than the tolerance, T. The ii segmented background image corresponds to the image of the background 311 with the object 130 removed. If needed, the remaining pixels in the segmented background image those corresponding to the removed object pixel locations) can be assigned any desired value and/or further processed using known image processing techniques.
If only the image of the object 130 is desired, steps 544 to obtain 311 need not be performed. Similarly, if only the image of the background 312 is desired, steps 542 and 130 -37- Y0994056: need not be performed.
In an alternative preferred embodiment, a translucent part of the object 131 (for example, a plastic cover) may be separated from an opaque part of the object 131, L_ adding i'^ steps 552, 554, and 556. In this embodiment, branch 542 goes to step 552 instead of step 540. Before step 552, it has been determined already that the pixel in the first image is brighter than its corresponding pixel in the second image. Step 552 determines if the object 130 pixels of the second image (the object 131 under low illumination) are brighter than a value V, a sec- *l ond tolerance value. If so, branch 553 is taken and the object pixel belongs to the translucent part 554 of object 130. (The object is translucent at this pixel location since some ambient light passed through the object 130 and 9 5 was imaged at this location when the light 110 was switched to low illumination. If not, then branch 555 is taken and the pixel belongs to opaque part 556 of object 130. (No ambient light, or an amount below the tolerance, V, is meas- -38- Y0994C06 ured through the opaque part of the object 130.) The value V may be constant f or each pixel in, the second image. or may be variable, depending, for example, on the position on the pixel in the second image. -Note that 'the value, V, may further be computed as describe above, from an image of the background 135 alone, by choosing a V such that 95% to of the background image is brighter than V. A preferred value for V is 20% of the brightest image intensity.
In step 554, a translucent object image is created. In this step, each pixel in the first image (which belongs to the object) which corresponds to a pixel in the second image that is brighter than the value V, corresponds a translucent part of object 130 and is stored in a translucent object im- .0*2z, age. After al~l pixels of the first and second images are so processed, the translucent object image will contain only the image of the translucent parts of object 130. if needed, the remaining pixels of the translucent object image may be assigned any desired valae and/or processed further.
In step 556, an opaque object image is created. In this step, c-ach pixel in the first image (which belongs to -the -39- 1 1 Y0994056 object) which corresponds to a pixel in the second image equal to or darker than the value V, corresponds to an opaque part of object image 130 and is stored in the opaque object image 556. After all pixels of the first and secon images are so processed, the opaqu object image will coon tain only the image of the opaque parts of object 130. If needed, the remaining pixels of the opaque object image may be assigned any desired value and/or be further processed.
If only an image of the opaque parts of the object 130 is desired, step 554 need not be performed. Similarly, if only an image of the translucent parts of the object 130 is desired, step 556 need not be performed.
In another preferred embodiment, step 552 is combined with: I step 540 and steps 542 and 130 are removed. This results in S the translucent object image or the opaque object image (or, both) but not the complete segmented object image 130.
Other combinations of steps 552, 554 and 556 with the previous steps are within the contemplation of this invention.
Y0994056 After the image is segmented 220, a computation of one or 'i more target object features is performed. Refer to step 230' of Figure 2. The computation 230 is performed by the computer 140 and is used to determine features of the target object. This determination is made by novelly performing this step 230 only on the separated out (segmented) image 130 of the target object obtained in step 220. For each pixel in the segmented object image, features are determined. For example, such features can be computed using the color of a single pixel, or using the (color) value of a pixel and the (color) values of its surrounding pixels.
Features include, but are not limited to, color, shape, texture, density of the segmented image of target object.
Normally, the feature(s) are represented by one or more feature values.
Once one or more features are determined 230, the feature or set of features is characterized 240. Histogramming is a preferred way of doing the characterization 240. See the description of Figure 6 below. However, other methods of characterizing feature(s) can be used. For example, median feature value, first order (mean value) and/or higher order -41- Y0994056 statistics of computed feature values, or any statistic that can be derived from the computed set of feature values can be used. Given this disclosure, one skilled in the art could develop other equivalent ways to characterize features.
The normalization step 250 of the algorithm 200 is a novel step for making the characterized feature(s) of an object independent of the size of the actual object 131 being imaged. This step also enables one or multiple instances of object 131 to be recognized by the apparatus 100 independent of the number of objects 131, or size of objects 131, in the scene. Normalization 250 is performed on one or more of the 0. computed feature characterization(s). A preferred method of normalization can be done with respect to area or length, obtained by counting number of pixels in segmented ob-: ject image (see the description of Figure 7, below) or by counting number of pixels on boundary of segmented object image (see the description of Figure 12, below).
0 Other methods of normalization, normalizing with respect to any other characteristic derived from segmented ob- -42- Y0994056 ject image, are also within the contemplation of the invention.
Another novel feature of the present invention enables the system 100 to be trained. If a normalized characterization of an object 131 is not recognized, not matched with reference information (step 260), the normalized characterization is checked 251 if it satisfies some storage criteria 255. If the normalized characterization of the unrecognized object meets the storage criteria 255', it will be stored 270 along with the other reference information. Therefore, the next time this object 131 is imaged by the system 100, it will be matched to a reference image and recognized. See 1. the description of Figure 9 below. Training allows the systemn 100 to be able to recognize objects that the system is not "hard-wired" (pre-programmed) to recognize, thus making the system more flexible. The stored characterization is 1 normalized so that the number of objects 131 used for refer- 2. ence characterization can be different from number of objects 131 used for developing target characterization.
-43- SY0 994056 Storage criteria can include any criteria established by the system 100 design. The ability to select and/or create i storage criteria 255 can also be given to a user through an interface 160. A simple storage criteria might be to store any information provided about an unrecognized object in the reference database 270. Other storage criteria might include, but are not limited to: the quality of image 210 is good; a large percentage of target object occupies image 210; characterizations should be sufficiently close (in the sense of 840 described in Figure 8) to refer- :i,3 ences of target object in database.
In step 260 of the algorithm 200, normalized characteristics l of the target object 131 are compared 260 to one or more S"a' normalized reference object characteristics. This compar-.
ison 260 depends very much on the method for characterizing object features, for which examples are given in step 240 above. One preferred comparison 260 is done with of area or length normalized histograms.
One or more reference object characteristics are stored 270 on a memory storage device. This device can be located in -44- Y099 4056 memory on the computer 140 or a separate secondary storage device 144. A preferred method for storing 270 the reference object characteristics is to use a series of area normalized feature histograms that characterized object features. Each of these series of area normalized histograms has associated with it a unique object type identifier. A preferred method of storing the area normalized feature histograms is by using a vector of normalized feature histograms. That is, the normalized frequencies of occurrence of the different feature values.
Figure 6 is a flow chart showing one preferred method of developing a histogram of a feature. In this non-limiting ex- S ample, the feature, Fl, Hue is used. However, any feature S**1 that can be extracted from the segmented image can be used.
Note that the present invention novelly extracts the feature S only from the segmented object image(s) 130. The feature histogrammed also can be derived from other information about the segmented object. For example, Hue could be derived from some other feature in a color map.
To develop the histogram 650, the scene is first imaged Y" 0994056 (step 210 of Figure 2) and the object image 130 is segmented 220. The feature to be histogrammed is then computed or de- termined using prior art techniques 230, preferably on a pixel by pixel basis (but could also be done for every other pixel, or any pre-determined subset of pixels) Prior art techniques are then used to develop 640 the histogram 650 of the feature, Fl. For example, a histogram array of M intervals (bins) is 1 0 first initialized to zero, Then, on a pixel by pixel basis, the Hue of pixel is computed. This computed Hue value for a particular pixel is quantized so that it will fall into one S, of the M bins, say Bin(x), of the histogram. The content of Bin(x) is then incremented by one, New Bin(x) Old Bin(x) 1. This is done for all pixels in segmented object image 130, or for selected subsets every other one) of these pixels.
The Hue histogram 650 is a representation of how color (Hue) 20 is distributed in the image of segmented object(s) 130. In other words, the content of each bin describes how many pixels in 130 have color represented by that bin. If Fl is -46- 1 Y0994056 some other feature, it is a representation of how that fea- ture is distributed in image of object 130. The content of each bin describes how many pixels in 130 have feature value: represented by that bin.
Figure 7 is a flow chart showing the steps of normalizing a histogram feature and how these normalized feature histograms are unaffected by the size or number of the object(s) 131 imaged..
Image 320 is a segmented color image exhibiting one segmented object 131 image 130. Image 720 is a segmented color image of three instances of object 131, exhibiting three i segmented object images 130. One or more feature(s) F1l are computed as described in Figure 6, and two histograms are developed, histogram 745 and histogram 740, respectively.
In step 750, each histogram (745, 740) is normalized using the same method of normalization 750. Since the present invention novelly normalizes 750 only the segmented images of the objects (130) in each image (320, 720), the resulting normalized histogram (770 and 760 respectively) are identical. This result occurs even though the image 720 with a -47- '7 i Y0994056 larger number of objects 131 will contribute a higher pixel count to each bin of the histogram 740 than the image 320 with a fewer number of objects 131 will contribute to its respective histogram 745. (Note that the same effect occurs if the size of the object 131 is greater 720 in one image than in the other 320.) For example, area normalizing creates approximately equal normalized histograms (760, 770) because the contribution of the segmented image to its histogram is divided by its respective image area.
:r i~i- i~:is;a ,too 99 *o 9 9 9 9*9* eoOr .9 ee o 9 .*15 9.
be 9 a 9 a* 9 9 9 Areal, that is the segmented object image area 130 in color image 320 is computed by adding the content of all the bins of histogram 745. Area2, that is the segmented area for all the objects 130 (or the larger sized object) for image 720 is computed in same fashion. To obtain area normalized histogram 760, histogram 745 is divided, bin by bin, by the value Areal. The area normalized histogram 770 is computed by dividing bin by bin histogram 740 by Area2. After this operation, area normalized histogram 760 is approximately equal to area normalized histogram 770 and readily compared 260 as in the description of Figure 8.
-48- Y0994056 Normalization can be done with respect to any property that can be extracted from segmented object image 130. Area, length, and size are examples. Other measures that describe the shape can be used, such measures include but are not limited to, second and higher-order (shape) moments, the I- 1 size of bounding rectangle, area of the convex hull of object image 130.
Figure 8 illustrates the step 260 of algorithm 200 that compares 840 the normalized characterizations (760, 770) of the segmented target image 130 to one or more stored normalized reference characterizations 270. Characterization 810 represents normalized characterization of some segmented image o*oo" containing a target object. This characterization is obtained as described in Figure 7. Block 820 is a list (database) of area normalized reference characterizations S obtained as described, in Figure 9. These are representations of the objects that the system is to be able to i recognize. Each of the plurality of normalized characterization representations are labeled typically as 831, 837. Only six are shown, but the number of area normalized histogram representations can be very large,, in the -49- Y0994056 100s or even 100Q0s. Each object to be recognized should be i i represented by at least one normalized characterization but 1 can be represented by more than one normalized characteriza- tion. Each area normalized characterization in 820 has associated with it a descriptive identifier of the object that the normalized characterization is developed from. Characterization 810 and reference characterizations 820 are not limited to one characterization, representation can be mul- tiple characterizations. In that case, multiple characterizations are developed from the image of the target object while multiple characterizations represent each reference object. Again, each such collection of characterizations is o associated with a unique object identifier. See the description of Figure Block 840 shows the comparison/matching of the target characterization to the reference characterizations. A preferred means of matching/comparing characterizations is to S e determine a distance measure, LI, between target histogram and reference histograms. For example, let target histogram 810 be represented as a vector T of numbers and reference histograms 820 as vectors Rl through some RN. For this dis-
Y0994056 closure, the best match of the target histogram T is defined as that reference histogram RI for which the L1 distance (sometimes called Manhattan distance) between T and R1 RI RN is smallest. That is, RI would give the smallest LI distance of distances Dist J=1,2 N Matching algorithms like this are well known as nearest neighbor classification. Any measure of distance that exhibits the usual properties of a distance measure (prior art) can be used here. Further, other measures that do not exhibit properties of distance, Histogram Intersection, could be used. Weights can be associated with the components of target histogram T and reference histograms R S RN, resulting in a component-wise weighted distance measure.
If target object and reference objects are represented by multiple histograms, preferred representations can be viewed as higher dimensional vectors containing multiple concatenated histograms, T' and Rl' RN'. One preferred way to define the best match of such a concatenated target histogram to T' is defined as that concatenated reference -51- Y0994056 histogram RI' for which the L1 distance between T' and RI' RN' is smallest. Here, different weights may be as- signed to different subvectors, representing different feature histograms, in the L1 distance. Again, any distance measure can be used, and also measures that do not exhibit properties of distance, Histogram Intersection, can be applied. Weights can be associated with every with the components of target histogram T' and reference histograms Rl' RN', resulting in a icomponent-wise weighted distance measure.
It is intended in this invention that object(s) 131 shown is *ag.
.of one type. Presenting multiple objects, Object A and Object B, can result in unpredictable results. ,The most .15 likely result is that objects are flagged as unrecognizable.
It could happen however, due to the distance measure used, that recognition result is: object is Object A; object is Object B; object is Object A or Object B presented as choices in user interface of Figure 5. The latter happens when mixed objects are of similar appearance, like, Granny Smith apples and Golden Delicious apples. It is unlikely that objects are recognized as some other Object C -52- Y0994056 stored in 820.
Figure 9 is a flow chart showing the method 910 s for training the system by adding to storage 270 (concatenated) reference histograms that meet certain storage criteria 255.
The traning method 910 enables the apparatus 100 to recognize new objects/items, those not originally stored in the system storage 270. The training 910 begins by presenting the apparatus with an image 920 of the object/item. The image is segmented 220'and then features are determined 230 for histogramming 640 'as described above. The normalized (concatenated) histogram 750 is compared 260 as before. If the target normalized histogram is matched with a reference
normalized histogram in storage, the target image is recognized. If not, the method 910 continues to check the target normalized image against certain storage criteria 255. If the storage criteria are not met, the method ends 940. If the target normalized image meets the storage criteria 255,
the target normalized image is stored in the storage device 270 where it can be used later to match other target images.
It is important that image 920 is obtained with device oper- -53- Y0994056 ating in same fashion as will be used later on recognize different instances of said object(s) 131. The preferred embodiment of such a device is described in Figure 4 with and polarized light through filter 410 and polarizing filter 420 on camera. Polarization is of particular importance because during training and recognition, object(s) 131 can have very unpredictable specular reflection (glare) effects.
Training also can be achieved through interactive input/output device 160; it can be achieved through human intervehtion, either by cashier in front end or by produce manager in back room. This is further described in Figure Figure 10 is a flow chart showing the steps of extracting more than one feature from an image and using more than one feature to identify an object. The method begins with an image 320 of an object 130 and background 311. As before, the object image 130 is segmented 220 from the background 311. Multiple features are then extracted from segmented image 320 in the same manner as Hue Fl (230) is extracted as described above. Blocks 1010 1020 typically refer to -54- Y0994056 other plurality of features extracted. These include, but are not limited to, Saturation, Intensity, Texture (described in Figure 11), Boundary Shape (described in Figure 12) and Density (described in Figure 13). As for color, the HSI representation is the preferred means of representing color for this invention. Other color representations may be used, including, but not limited-to RGB, Munsell, opponent colors.
i0 Once the features Fl FN are extracted, they are L histogrammed and normalized as described above. Although VS many features, like color can be area normalized, other nor- S, malizations length, boundaries) are possible that S might be particularly suited to a feature. For example, see .below in Figure 12 for shape histograms In step 840, each of the extracted features, Fl FN, are i compared. This comparison is already described in Figure 8.
Essentially the extracted N histograms (of features F1 FN) are concatenated in a long histogram and comparison is based on some distance measure between target concatenated histogram and reference concatenated histograms. In this Y0994056 distance measure, histograms of individual features Fl EN could be weighted with different weights wl wN. Alternatively, in distance measure, each individual component of concatenated histogram can have a individual weight. As mentioned above, features Fl FN include, but are not lim- ited, Hue, Saturation, Intensity, Texture, Shape. Not all of these features have equal discriminative power, and therefore, weights wl wN may be used. Moreover, not every component of an individual feature, say FI, may have equal discriminative power. So, individual feature histograms can be weighted differently component-wise.
,Figure 11 illustrates how texture is used as a normalized Srecognition feature. Block 210 is an image of an object/item that exhibits the feature of texture 1120. As before, the object image is segmented 220 from its back- S, ground. In step 1140 the texture feature is computed from the segmented image. Any texture measure known in the prior art could be used in this 1140 computation. However, two novel computation are preferred.
The first preferred means of texture computation is a novel -56- Y0994056 texture measure A: Segmented image is transferred into a binary image by selecting a threshold Tb using methods in the prior art. If image brightness is larger than Tb, binary image is set to 1; if image brightness is smaller than Tb, binary image is set to 0. Other means, known to a person skilled in the art, for binarizing images also can be used. The result is blob-like black and white image. Each blob can be charac- ,ried by a Width and a Length; the texture measure (WE) associated with a blob is given by Width Eccentricity Width/Length SThis is a texture measure which is determined using region calculations.
r A second preferred novel texture measure B is the following.
The image is convolved using prior art methods with 2 mask, this convolution is performed both vertically and horizontally, denoted by Vconv and Hconv, respectively. At -57- Y0994056 each pixel x where the convolution result is over some threshold T2, the vector consisting of Magnitude and Direc- j i tion Magnitude sqrt Vconv(x)**2 Hconv(x)**2) Direction arctan Vconv(x)/Hconv(x) are defined as the texture measure. This is a texture measure which is determined using edge calculations.
The texture feature can also be histogrammed in a novel way 1150, that is, only over segmented object image 1120. The Sl texture measures are histogrammed over the segmented image as described above. For texture measure A, this results in
(Width-Eccentricity) histogram, where Width and Eccentricity 15 are defined above. Texture measure B, gives a (Magnitude- Direction) histogram, where Magnitude and Direction are defined above. For the Direction histogram, the maximum direction in the Direction histogram is computed and the histogram cyclically shifted to bring this peak to the center. This will make the Direction histogram independent of the rotation under which texture is imaged.
-58- Y0994056- Texture histograms are normalized by count. Here count can be each pixel in segmented object image 1120, or count could i be those pixels in segmented object image 1120 that actually exhibits texture. Other shape measures extracted from the textured region can be envisioned by a person skilled in the art. A resulting normalized texture histogram is shown as 1170. Use of texture histograms for recognizing objects is believed novel.
Figure 12 is a flow chart showing the steps of using shape as a recognition feature. The image 210 is segmented 220 as above. Next a determination of which pixels in object image 130 are boundary pixels is made 1210. A pixel P is a boundary pixel if one or more neighboring pixels of P belong to i background image 311. Next a determination is made of a bou-ndary shape property 1220 for each boundary pixel P. A preferred shape measure used by this invention is local boundary curvature at pixel P. The radius R of a circle that fits the center pixel P and number of the surrounding boundary pixels is first computed by computer 140. The curvature I/R describes the local degree of variation for pixel P zero curvature for a straight line boundary, high cur- -59- Y0994056 vature for a locally "wiggly" boundary. An apple will have a roughly constant curvature boundary, while a cucumber has a piece of low curvature, a piece of low negative curvature, and two pieces of high curvature (the end points). Other shape measures are possible.
The boundary shape feature(s) are then histogrammed 1230 The histogramming is developed by the computed shape properties of boundary pixels P. Instead of over an area, histograms here are developed from a collection of image pixels P that comprise the boundary of object image 130.
The normalization done 1235 is a length normalization of the S* shape histogram. Bin by bin, the histogram of 1230 is divided by the total number of boundary pixels P. The result
is that the length normalized shape histogram of one object 00 .0° image 130 is equal to the length normalized shape histogram of multiple object images 130. Length normalized object image boundary shape histograms are a novel feature of this invention. Other normalizations related to length of the object image boundary are possible.
Y0994056 Density can be an important recognition feature. A pound of white onions weighs as much as a pound of mushrooms, but the volume of the mushrooms is much larger than that of the white onions. Therefore, the relation between weight and volume is important. This relation is object density determined by Density Weight (Object 131) Volume (Object 131)
5 a Figure 13 is a block diagram showing the computer 140 connected to'a weighing device 170 that determines the weight of the object(s) 131. To use weight as recognition feature, device 170 reports the weight of object(s) 131 to computing device 140. In a preferred embodiment, the system 100 uses S a weight a "Density" feature defined as onsity" Weight (Object 130) Area (Segmented object 131) This measure does not embody the conventional means of referring to density, rather it is a measure of pressure. It is called the average projected density.
Y0994056 True density of object(s) 131 can only be computed very roughly. To get an idea of the volume of object(s) 131, the boundary contour(s) of 130 can be approximated with an ellipse and the volume(s) of 131 can be approximated by the volume of an ellipsoid of revolution developed from the approximated ellipse. Density, then, is given by Weight/Volume.
Other means for estimating volume from a projected segmented object image 130 are within the scope of this invention.
*b I Multiple reference histograms representing the same feature Hue) can be used to recognize a given object.
Figure 14 shows an image 1405 where segmented object image 130 has two distinct regions, the leaves 1410 and the grapes 1420. The image 1405 comprises the object 130 and S' background 311. The object image 130 is segmented along with its first object region 1410 and its second 1420 object a* a region. These object regions are recognized and defined by using a segmentation algorithm. A preferred algorithm is the use of an area normalized Hue histogram for detecting if there are two or more distinct peaks.
-62- SY0994056 These regions are histogrammed and area normalized separately. Area normalized histograms 1450 and 1455 correspond to the first 1410 and second 1420 region, respectively, and are compared to reference histograms as described above.
Additionally, relative location 1430 of regions 1410 and 1420 can be taken into account during matching (Figure 8).
This part of the invention accounts for items where a feature, color, but not limited to color, is not uniform over the surface of the object 131 and hence not uniform over the segmented object image 130. A typical example are carrots with the leafy part left on.
s 'Figure 15 shows an optional human interface 160. It comtlmoo prises a preferred means of displaying 164 of pictorial (or otherwise explained) description(s) 1510, 1520, 1530 and 44 1540 of various possible identities of object(s) 131 that are determined by apparatus 100. In most cases, object(s) 131 can be uniquely identified by comparison 260 to the reference database 270. However, in some case, there may be a match to more than one reference histogram, the target object normalized histogram may be approximately the same as more than one reference histogram. In these cases a human -63- Y0994056 can be novelly asked through the interface 160 to make the final recognition decision. A preferred embodiment of the interface 160 offers four or fewer choices 1510, 1520, 1530, 1540. More choices can be optionally requested as explained later. The human can communicate th'e decision to computer 140 through any means, touch, voice, mouse, keyboard. In addition, a means (button) 162 can be provided on the interface to enable the user to determine when and if a histogram should be added to the reference object database 820 in storage 270, if the system is to be trained with that data to recognize (or better recognize) instance Sof object 131 when presented to system 100 at some future point.
.:1S Figure 16 shows an interface 160 with a function, like a tt B browsing key, that allows the user to browse for an object Sidentity. A browsing key refers to a key word or key feature by which to narrow down the human guided search for object identity in the database 820. Examples of such keys are, but are not limited to: Red 1612, Green 1613, Yellow 1614, Brown 1615, Round 1616, Straight 1617, Leafy 1618, Apples 1619, Citrus Fruits 1620, Peppers 1621, and Potatoes -64- Y0994056 1622, as displayed in 1610. The user can communicate through touch, voice, mouse, keyboard, etc. The key 1600 will respond with either another instance of 1610, in which the choices presented 1612 1622 are more specific, or with screen 1630 where a final decision can be made. If 1619, e.g. apples, is selected, 1600 will present human with screen 1630, offering descriptions (sketches, photographs, words) 1631 1641 of identity of the object(s) 131. The iser can select choices on the screens using various known input devices. Any other human-friendly method or means can be used.
Figure 17 is a block diagram showing optional apparatus used with system 100 to price objects. A weighing device 170 is
used to determine the weight of object(s) 131. The appara- *e tus 100 recognizes the object as described above. Once the object is recognized, a price of the object is determined.
The weight 170 and or the count (number of items present) of the object is used if required in the pricing. The prices of the objects are stored in memory 144.
Y0994056 Price device 1710 is attached to apparatus 100 to communi- cate the price to the user. Price device 1710 can be a printing device, display device, or any other means of cornmunicating the price of the object. The price can also be displayed on the interactive output device 160. If price is specified by pound, the computer 140 calculates price as Price Weight (Price of object 131 per pound) If price is specified by count, computer 140 calculates price as ij !Price Count (Unit price of object 131) o* o Item count can either be obtained through human intervention or can be estimated.' For entering count through human intervention, system 100 Swill simply prompt human to enter count if item 131 is indicated in computer memory 140 as being sold by count lemons, limes). There are two ways automatically estimating Count: a) Apparatus 100 has average weight of object 131 in memory 144 and Count is computed as -66- Y0994056 Count =Weight /Average weight (object 131) after identity of object 131 is established from segmented object image 13.
b) Apparatus 100 makes an estimate of number of segmented object images 130 are present, and Count =Number of segmented object images 130.
Priority Applications (2)
|Application Number||Priority Date||Filing Date||Title|
|US08/235,834 US5546475A (en)||1994-04-29||1994-04-29||Produce recognition system|
|Publication Number||Publication Date|
|AU1503495A AU1503495A (en)||1995-11-09|
|AU685904B2 true AU685904B2 (en)||1998-01-29|
Family Applications (1)
|Application Number||Title||Priority Date||Filing Date|
|AU15034/95A Expired - Fee Related AU685904B2 (en)||1994-04-29||1995-03-23||Produce recognition system|
Country Status (10)
|US (1)||US5546475A (en)|
|EP (1)||EP0685814B1 (en)|
|JP (1)||JP3369772B2 (en)|
|KR (1)||KR100190737B1 (en)|
|CN (1)||CN1092375C (en)|
|AT (1)||AT196695T (en)|
|AU (1)||AU685904B2 (en)|
|CA (1)||CA2144404C (en)|
|DE (1)||DE69518947T2 (en)|
|ES (1)||ES2151000T3 (en)|
Families Citing this family (191)
|Publication number||Priority date||Publication date||Assignee||Title|
|US7904187B2 (en)||1999-02-01||2011-03-08||Hoffberg Steven M||Internet appliance system and method|
|US8352400B2 (en)||1991-12-23||2013-01-08||Hoffberg Steven M||Adaptive pattern recognition based controller apparatus and method and human-factored interface therefore|
|US10361802B1 (en)||1999-02-01||2019-07-23||Blanding Hovenweep, Llc||Adaptive pattern recognition based control system and method|
|US5933524A (en) *||1993-09-27||1999-08-03||Siemens Aktiengesellschaft||Method for segmentation of digital color images|
|JP3468877B2 (en) *||1994-10-27||2003-11-17||矢崎総業株式会社||Plant automatic diagnosis method and apparatus|
|US5649070A (en) *||1995-02-17||1997-07-15||International Business Machines Corporation||Learning system with prototype replacement|
|EP0727760A3 (en) *||1995-02-17||1997-01-29||Ibm||Produce size recognition system|
|US5649021A (en) *||1995-06-07||1997-07-15||David Sarnoff Research Center, Inc.||Method and system for object detection for instrument control|
|US6069696A (en) *||1995-06-08||2000-05-30||Psc Scanning, Inc.||Object recognition system and method|
|US6959870B2 (en) *||1999-06-07||2005-11-01||Metrologic Instruments, Inc.||Planar LED-based illumination array (PLIA) chips|
|US5839104A (en) *||1996-02-20||1998-11-17||Ncr Corporation||Point-of-sale system having speech entry and item recognition support system|
|US7650015B2 (en) *||1997-07-22||2010-01-19||Image Processing Technologies. LLC||Image processing method|
|US5969317A (en) *||1996-11-13||1999-10-19||Ncr Corporation||Price determination system and method using digitized gray-scale image recognition and price-lookup files|
|US5933502A (en) *||1996-12-20||1999-08-03||Intel Corporation||Method and apparatus for enhancing the integrity of visual authentication|
|FR2757974B1 (en) *||1996-12-27||1999-02-12||Sagem||Optical fingerprint sensor|
|AT217717T (en) *||1996-12-27||2002-06-15||Sagem||Optical fingerprint sensor|
|FR2757975B1 (en) *||1997-05-21||1999-02-12||Sagem||Optical fingerprint sensor|
|FR2763719B1 (en) *||1997-05-21||1999-07-30||Sagem||Optical fingerprint sensor|
|FR2763720B1 (en) *||1997-05-21||1999-07-16||Sagem||Optical fingerprint light fingerprint sensor|
|US7212654B2 (en) *||1997-06-20||2007-05-01||Dawn Foods, Inc.||Measurement of fruit particles|
|US6504957B2 (en)||1997-07-07||2003-01-07||General Electric Company||Method and apparatus for image registration|
|US6219438B1 (en) *||1997-09-02||2001-04-17||Lucent Technologies Inc.||Produce indentifier using barcode scanner and wavelet image processing and having compensation for dirt accumulated on viewing window|
|US5924575A (en) *||1997-09-15||1999-07-20||General Electric Company||Method and apparatus for color-based sorting of titanium fragments|
|US6035055A (en) *||1997-11-03||2000-03-07||Hewlett-Packard Company||Digital image management system in a distributed data access network system|
|US6181817B1 (en)||1997-11-17||2001-01-30||Cornell Research Foundation, Inc.||Method and system for comparing data objects using joint histograms|
|JP5088990B2 (en) *||1999-01-26||2012-12-05||Ｈｏｙａ株式会社||Autofluorescence imaging system for endoscope|
|EP1717684A3 (en)||1998-01-26||2008-01-23||Fingerworks, Inc.||Method and apparatus for integrating manual input|
|US6424745B1 (en) *||1998-05-19||2002-07-23||Lucent Technologies Inc.||Method and apparatus for object recognition|
|US6334092B1 (en) *||1998-05-26||2001-12-25||Mitsui Mining & Smelting Co., Ltd.||Measurement device and measurement method for measuring internal quality of fruit or vegetable|
|US7417640B1 (en) *||1999-01-29||2008-08-26||Lg Electronics Inc.||Method for dominant color setting of video region and data structure and method of confidence measure extraction|
|US6157435A (en) *||1998-05-29||2000-12-05||Eastman Kodak Company||Image processing|
|US6363366B1 (en)||1998-08-31||2002-03-26||David L. Henty||Produce identification and pricing system for checkouts|
|US6332573B1 (en)||1998-11-10||2001-12-25||Ncr Corporation||Produce data collector and produce recognition system|
|US6155489A (en) *||1998-11-10||2000-12-05||Ncr Corporation||Item checkout device including a bar code data collector and a produce data collector|
|US6296186B1 (en) *||1998-11-19||2001-10-02||Ncr Corporation||Produce recognition system including a produce shape collector|
|US6624761B2 (en)||1998-12-11||2003-09-23||Realtime Data, Llc||Content independent data compression method and system|
|US6445812B1 (en) *||1999-01-22||2002-09-03||Siemens Corporate Research, Inc.||Illumination compensation system for industrial inspection|
|US6567797B1 (en)||1999-01-26||2003-05-20||Xerox Corporation||System and method for providing recommendations based on multi-modal user clusters|
|US6922699B2 (en) *||1999-01-26||2005-07-26||Xerox Corporation||System and method for quantitatively representing data objects in vector space|
|US6941321B2 (en)||1999-01-26||2005-09-06||Xerox Corporation||System and method for identifying similarities among objects in a collection|
|DE60045699D1 (en) *||1999-01-29||2011-04-21||Sony Corp||Data description method and data machining device|
|TW419634B (en) *||1999-02-02||2001-01-21||Ind Tech Res Inst||Automatic detection system and method using bar code positioning|
|US6336082B1 (en)||1999-03-05||2002-01-01||General Electric Company||Method for automatic screening of abnormalities|
|US6604158B1 (en)||1999-03-11||2003-08-05||Realtime Data, Llc||System and methods for accelerated data storage and retrieval|
|US6601104B1 (en)||1999-03-11||2003-07-29||Realtime Data Llc||System and methods for accelerated data storage and retrieval|
|EP1041378A1 (en) *||1999-03-29||2000-10-04||Ncr International Inc.||Produce recognition system including a produce data collector|
|US6603877B1 (en) *||1999-06-01||2003-08-05||Beltronics, Inc.||Method of and apparatus for optical imaging inspection of multi-material objects and the like|
|US8042740B2 (en) *||2000-11-24||2011-10-25||Metrologic Instruments, Inc.||Method of reading bar code symbols on objects at a point-of-sale station by passing said objects through a complex of stationary coplanar illumination and imaging planes projected into a 3D imaging volume|
|US6260023B1 (en) *||1999-06-14||2001-07-10||Ncr Corporation||Transaction processing system including a networked produce recognition system|
|US6431446B1 (en)||1999-07-28||2002-08-13||Ncr Corporation||Produce recognition system and method|
|US6529855B1 (en) *||1999-07-28||2003-03-04||Ncr Corporation||Produce recognition system and method|
|AUPQ212499A0 (en) *||1999-08-10||1999-09-02||Ajax Cooke Pty Ltd||Item recognition method and apparatus|
|US6748457B2 (en)||2000-02-03||2004-06-08||Realtime Data, Llc||Data storewidth accelerator|
|US6446869B1 (en)||2000-02-10||2002-09-10||Ncr Corporation||Ambient light blocking apparatus for a produce recognition system|
|US6501547B1 (en)||2000-02-18||2002-12-31||Ncr Corporation||Hand-held produce recognition system and produce data collector|
|US6505775B1 (en)||2000-04-25||2003-01-14||Ncr Corporation||Produce data collector with enhanced LVF spectrometer|
|US6471125B1 (en)||2000-05-24||2002-10-29||Ncr Corporation||Method of tracking produce selection data|
|US6530521B1 (en)||2000-07-17||2003-03-11||Ncr Corporation||Produce recognition apparatus and method of obtaining information about produce items|
|US6409085B1 (en)||2000-08-16||2002-06-25||Ncr Corporation||Method of recognizing produce items using checkout frequency|
|US6606579B1 (en) *||2000-08-16||2003-08-12||Ncr Corporation||Method of combining spectral data with non-spectral data in a produce recognition system|
|US6601767B1 (en)||2000-08-16||2003-08-05||Ncr Corporation||Ambient light sensing apparatus and method for a produce data collector|
|US6658138B1 (en)||2000-08-16||2003-12-02||Ncr Corporation||Produce texture data collecting apparatus and method|
|US6412694B1 (en)||2000-09-20||2002-07-02||Ncr Corporation||Produce recognition system and method including weighted rankings|
|US6668078B1 (en)||2000-09-29||2003-12-23||International Business Machines Corporation||System and method for segmentation of images of objects that are occluded by a semi-transparent material|
|US8692695B2 (en)||2000-10-03||2014-04-08||Realtime Data, Llc||Methods for encoding and decoding data|
|US9143546B2 (en)||2000-10-03||2015-09-22||Realtime Data Llc||System and method for data feed acceleration and encryption|
|US7417568B2 (en)||2000-10-03||2008-08-26||Realtime Data Llc||System and method for data feed acceleration and encryption|
|US6577983B1 (en)||2000-10-06||2003-06-10||Ncr Corporation||Produce recognition method|
|US6510994B1 (en)||2000-10-06||2003-01-28||Ncr Corporation||Triggering method for a produce recognition system|
|US7595878B2 (en) *||2000-10-13||2009-09-29||Chemimage Corporation||Spectroscopic methods for component particle analysis|
|US7890386B1 (en) *||2000-10-27||2011-02-15||Palisades Technology, Llc||Method for use with a wireless communication device for facilitating tasks using images and selections|
|US7841533B2 (en)||2003-11-13||2010-11-30||Metrologic Instruments, Inc.||Method of capturing and processing digital images of an object within the field of view (FOV) of a hand-supportable digitial image capture and processing system|
|US7221794B1 (en) *||2000-12-18||2007-05-22||Sportsvision, Inc.||Foreground detection|
|US7386046B2 (en)||2001-02-13||2008-06-10||Realtime Data Llc||Bandwidth sensitive data compression and decompression|
|WO2002086651A2 (en) *||2001-04-23||2002-10-31||Koninklijke Philips Electronics N.V.||Method of controlling an apparatus|
|US6947575B2 (en) *||2001-05-24||2005-09-20||Trw Inc.||Apparatus and method for determining vehicle occupant characteristic utilizing imaging with provided light|
|DE10156157A1 (en) *||2001-11-15||2003-05-28||Bsh Bosch Siemens Hausgeraete||Method and device for identifying an object|
|GB0130802D0 (en) *||2001-12-22||2002-02-06||Koninkl Philips Electronics Nv||Description generation|
|US7254268B2 (en) *||2002-04-11||2007-08-07||Arcsoft, Inc.||Object extraction|
|US9955551B2 (en) *||2002-07-12||2018-04-24||Yechezkal Evan Spero||Detector controlled illuminating system|
|WO2004038669A1 (en) *||2002-10-22||2004-05-06||Umc Utrecht Holding B.V.||System for remote transfer of a monitoring signal|
|US20040169656A1 (en) *||2002-11-15||2004-09-02||David Piponi Daniele Paolo||Method for motion simulation of an articulated figure using animation input|
|DE10326495B8 (en) *||2003-06-10||2004-12-16||Deutsche Post Ag||Method for processing mailpieces|
|US7496228B2 (en) *||2003-06-13||2009-02-24||Landwehr Val R||Method and system for detecting and classifying objects in images, such as insects and other arthropods|
|US7118026B2 (en)||2003-06-26||2006-10-10||International Business Machines Corporation||Apparatus, method, and system for positively identifying an item|
|US9614772B1 (en)||2003-10-20||2017-04-04||F5 Networks, Inc.||System and method for directing network traffic in tunneling applications|
|US20050166147A1 (en) *||2004-01-26||2005-07-28||Stein Russell L.||Produce Ripeness Enhancement|
|US7650835B2 (en) *||2004-01-26||2010-01-26||Russ Stein||Produce ripening system|
|JP4741804B2 (en) *||2004-02-25||2011-08-10||株式会社エヌ・ティ・ティ・ドコモ||Image processing apparatus and image processing method|
|US20050254728A1 (en) *||2004-05-13||2005-11-17||Zhuo-Ya Wang||Automatic cutting method for digital images|
|JP4648660B2 (en) *||2004-07-21||2011-03-09||大日本スクリーン製造株式会社||Acquisition of surface area arrangement of objects by image segmentation|
|US8024483B1 (en)||2004-10-01||2011-09-20||F5 Networks, Inc.||Selective compression for network connections|
|JP4327754B2 (en) *||2005-03-16||2009-09-09||富士フイルム株式会社||Image list generation apparatus and method, and program|
|JP2006350806A (en) *||2005-06-17||2006-12-28||Toshiba Tec Corp||Radio tag recognizing device and article sales data processor|
|US7756341B2 (en)||2005-06-30||2010-07-13||Xerox Corporation||Generic visual categorization method and system|
|FR2887987B1 (en) *||2005-06-30||2008-01-25||Sferis Sarl||Method and device for analyzing berries|
|US7783781B1 (en)||2005-08-05||2010-08-24||F5 Networks, Inc.||Adaptive compression|
|US8533308B1 (en)||2005-08-12||2013-09-10||F5 Networks, Inc.||Network traffic management through protocol-configurable transaction processing|
|US8275909B1 (en)||2005-12-07||2012-09-25||F5 Networks, Inc.||Adaptive compression|
|US7882084B1 (en)||2005-12-30||2011-02-01||F5 Networks, Inc.||Compression of data transmitted over a network|
|US8290275B2 (en) *||2006-01-20||2012-10-16||Kansai Paint Co., Ltd.||Effective pigment identification method, identification system, identification program, and recording medium therefor|
|US8565088B1 (en)||2006-02-01||2013-10-22||F5 Networks, Inc.||Selectively enabling packet concatenation based on a transaction boundary|
|US7873065B1 (en)||2006-02-01||2011-01-18||F5 Networks, Inc.||Selectively enabling network packet concatenation based on metrics|
|US8239268B2 (en)||2006-06-20||2012-08-07||Toshiba Tec Kabushiki Kaisha||Self-checkout terminal|
|NL1032435C2 (en) *||2006-09-05||2008-03-06||Maasland Nv||Device for automatically milking a dairy animal.|
|DE102006044365A1 (en) *||2006-09-20||2008-04-03||Mettler-Toledo (Albstadt) Gmbh||Automatic recognition device|
|US9356824B1 (en)||2006-09-29||2016-05-31||F5 Networks, Inc.||Transparently cached network resources|
|JP5213867B2 (en) *||2006-10-02||2013-06-19||イー・アイ・デュポン・ドウ・ヌムール・アンド・カンパニーＥ．Ｉ．Ｄｕ Ｐｏｎｔ Ｄｅ Ｎｅｍｏｕｒｓ Ａｎｄ Ｃｏｍｐａｎｙ||Method for measuring coating appearance and use thereof|
|IL179639D0 (en) *||2006-11-27||2007-05-15||Amit Technology Science & Medi||A method and system for diagnosing and treating a pest infested body|
|US8417833B1 (en)||2006-11-29||2013-04-09||F5 Networks, Inc.||Metacodec for optimizing network data compression based on comparison of write and read rates|
|US8130203B2 (en)||2007-01-03||2012-03-06||Apple Inc.||Multi-touch input discrimination|
|US8269727B2 (en)||2007-01-03||2012-09-18||Apple Inc.||Irregular input identification|
|US7855718B2 (en)||2007-01-03||2010-12-21||Apple Inc.||Multi-touch input discrimination|
|US9106606B1 (en)||2007-02-05||2015-08-11||F5 Networks, Inc.||Method, intermediate device and computer program code for maintaining persistency|
|US8794524B2 (en) *||2007-05-31||2014-08-05||Toshiba Global Commerce Solutions Holdings Corporation||Smart scanning system|
|US7988045B2 (en) *||2007-05-31||2011-08-02||International Business Machines Corporation||Portable device-based shopping checkout|
|US20090026270A1 (en) *||2007-07-24||2009-01-29||Connell Ii Jonathan H||Secure checkout system|
|US8544736B2 (en) *||2007-07-24||2013-10-01||International Business Machines Corporation||Item scanning system|
|US8472715B2 (en) *||2007-10-26||2013-06-25||Panasonic Corporation||Situation determining apparatus, situation determining method, situation determining program, abnormality determining apparatus, abnormality determining method, abnormality determining program, and congestion estimating apparatus|
|US8280763B2 (en) *||2008-02-26||2012-10-02||Connell Ii Jonathan H||Customer rewarding|
|US8746557B2 (en)||2008-02-26||2014-06-10||Toshiba Global Commerce Solutions Holding Corporation||Secure self-checkout|
|EP2250623A4 (en) *||2008-03-05||2011-03-23||Ebay Inc||Method and apparatus for image recognition services|
|US9495386B2 (en) *||2008-03-05||2016-11-15||Ebay Inc.||Identification of items depicted in images|
|US7889068B2 (en) *||2008-03-20||2011-02-15||International Business Machines Corporation||Alarm solution for securing shopping checkout|
|US8061603B2 (en) *||2008-03-20||2011-11-22||International Business Machines Corporation||Controlling shopper checkout throughput|
|US8229158B2 (en) *||2008-04-29||2012-07-24||International Business Machines Corporation||Method, system, and program product for determining a state of a shopping receptacle|
|US20090272801A1 (en) *||2008-04-30||2009-11-05||Connell Ii Jonathan H||Deterring checkout fraud|
|US20100053329A1 (en) *||2008-08-27||2010-03-04||Flickner Myron D||Exit security|
|US8346010B2 (en) *||2008-09-17||2013-01-01||Panasonic Corporation||Image processing device, imaging device, evaluation device, image processing method, and optical system evaluation method|
|US7991646B2 (en)||2008-10-30||2011-08-02||Ebay Inc.||Systems and methods for marketplace listings using a camera enabled mobile device|
|US8004576B2 (en) *||2008-10-31||2011-08-23||Digimarc Corporation||Histogram methods and systems for object recognition|
|US8113427B2 (en) *||2008-12-18||2012-02-14||Ncr Corporation||Methods and apparatus for automated product identification in point of sale applications|
|US8374801B2 (en) *||2009-01-09||2013-02-12||Chemimage Corporation||Automation of ingredient-specific particle sizing employing raman chemical imaging|
|US9208400B2 (en)||2009-02-13||2015-12-08||Cognitech, Inc.||Registration and comparison of three-dimensional objects|
|US8605989B2 (en) *||2009-02-13||2013-12-10||Cognitech, Inc.||Registration and comparison of three dimensional objects in facial imaging|
|US8825660B2 (en) *||2009-03-17||2014-09-02||Ebay Inc.||Image-based indexing in a network-based marketplace|
|CN101937505B (en) *||2009-07-03||2013-03-27||深圳泰山在线科技有限公司||Target detection method and equipment and used image acquisition device thereof|
|US9164577B2 (en)||2009-12-22||2015-10-20||Ebay Inc.||Augmented reality system, method, and apparatus for displaying an item image in a contextual environment|
|US8565538B2 (en) *||2010-03-16||2013-10-22||Honda Motor Co., Ltd.||Detecting and labeling places using runtime change-point detection|
|US8369617B2 (en) *||2010-04-12||2013-02-05||Nokia Corporation||Cross device image recognition improvement|
|CN101912847B (en) *||2010-08-02||2012-11-14||扬州福尔喜果蔬汁机械有限公司||Fruit grading system and method based on DSP machine vision|
|JP5544332B2 (en) *||2010-08-23||2014-07-09||東芝テック株式会社||Store system and program|
|JP5194149B2 (en) *||2010-08-23||2013-05-08||東芝テック株式会社||Store system and program|
|JP2012053708A (en) *||2010-09-01||2012-03-15||Toshiba Tec Corp||Store system, sales registration device and program|
|US9412050B2 (en) *||2010-10-12||2016-08-09||Ncr Corporation||Produce recognition method|
|US10127606B2 (en)||2010-10-13||2018-11-13||Ebay Inc.||Augmented reality system and method for visualizing an item|
|EA022480B1 (en) *||2010-12-23||2016-01-29||Тельман Аббас Оглы Алиев||Method of detection and identification of patterns and ornaments, and intelligent-information system for its implementation|
|JP5799593B2 (en)||2011-06-07||2015-10-28||株式会社寺岡精工||Product search device, product information processing device, and label issuing device|
|US9367770B2 (en)||2011-08-30||2016-06-14||Digimarc Corporation||Methods and arrangements for identifying objects|
|JP5431429B2 (en)||2011-09-06||2014-03-05||東芝テック株式会社||Information processing apparatus and program|
|EP2570967A1 (en) *||2011-09-13||2013-03-20||Nederlandse Organisatie voor toegepast -natuurwetenschappelijk onderzoek TNO||Semi-automatic check-out system and method|
|JP5194160B1 (en) *||2011-10-19||2013-05-08||東芝テック株式会社||Information processing apparatus, information processing method, and program|
|JP5551140B2 (en) *||2011-10-19||2014-07-16||東芝テック株式会社||Information processing apparatus and program|
|US9449342B2 (en)||2011-10-27||2016-09-20||Ebay Inc.||System and method for visualization of items in an environment using augmented reality|
|JP5551143B2 (en) *||2011-12-02||2014-07-16||東芝テック株式会社||Store system and program|
|US8897554B2 (en)||2011-12-13||2014-11-25||The Nielsen Company (Us), Llc||Video comparison using color histograms|
|US8897553B2 (en)||2011-12-13||2014-11-25||The Nielsen Company (Us), Llc||Image comparison using color histograms|
|US8750613B2 (en)||2011-12-13||2014-06-10||The Nielsen Company (Us), Llc||Detecting objects in images using color histograms|
|US9240059B2 (en)||2011-12-29||2016-01-19||Ebay Inc.||Personal augmented reality|
|JP5579202B2 (en) *||2012-01-16||2014-08-27||東芝テック株式会社||Information processing apparatus, store system, and program|
|JP5622756B2 (en) *||2012-01-30||2014-11-12||東芝テック株式会社||Product reading apparatus and product reading program|
|JP5586641B2 (en) *||2012-02-24||2014-09-10||東芝テック株式会社||Product reading apparatus and product reading program|
|JP5483629B2 (en) *||2012-02-29||2014-05-07||東芝テック株式会社||Information processing apparatus, store system, and program|
|US9881354B2 (en) *||2012-03-15||2018-01-30||Microsoft Technology Licensing, Llc||Image completion including automatic cropping|
|US9934522B2 (en)||2012-03-22||2018-04-03||Ebay Inc.||Systems and methods for batch- listing items stored offline on a mobile device|
|US9087269B2 (en) *||2012-08-24||2015-07-21||Google Inc.||Providing image search templates|
|JP5658720B2 (en) *||2012-09-06||2015-01-28||東芝テック株式会社||Information processing apparatus and program|
|US9224184B2 (en)||2012-10-21||2015-12-29||Digimarc Corporation||Methods and arrangements for identifying objects|
|ITMO20120266A1 (en) *||2012-10-31||2014-05-01||Charlotte Anna Maria Liedl||Device for the orientation of objects.|
|JP5936993B2 (en)||2012-11-08||2016-06-22||東芝テック株式会社||Product recognition apparatus and product recognition program|
|CN103028555A (en) *||2012-12-03||2013-04-10||天津理工大学||Raisin color sorting detection method based on HSI (Hue-Saturation-Intensity) color features|
|KR101490909B1 (en) *||2013-05-10||2015-02-06||현대자동차 주식회사||Apparatus and method for image processing of vehicle|
|CN103279763B (en) *||2013-05-25||2016-03-02||中北大学||A kind of wheel hub type automatic identifying method of structure based feature|
|JP5847117B2 (en) *||2013-05-28||2016-01-20||東芝テック株式会社||Recognition dictionary creation device and recognition dictionary creation program|
|JP5927147B2 (en)||2013-07-12||2016-05-25||東芝テック株式会社||Product recognition apparatus and product recognition program|
|JP6122805B2 (en) *||2013-07-16||2017-04-26||東芝テック株式会社||Information processing apparatus and program|
|CN104299337B (en) *||2013-07-16||2017-07-07||东芝泰格有限公司||Information processor and its control method|
|JP5826801B2 (en) *||2013-07-19||2015-12-02||東芝テック株式会社||Product recognition apparatus and product recognition program|
|CN103425969A (en) *||2013-08-07||2013-12-04||华南理工大学||Detection system and detection method for identifying type of wheel hub|
|US10650232B2 (en) *||2013-08-26||2020-05-12||Ncr Corporation||Produce and non-produce verification using hybrid scanner|
|US9177225B1 (en)||2014-07-03||2015-11-03||Oim Squared Inc.||Interactive content generation|
|JP6428240B2 (en) *||2014-12-17||2018-11-28||カシオ計算機株式会社||Product registration device, product recognition method, and program|
|JP6345612B2 (en) *||2015-02-06||2018-06-20||株式会社ソニー・インタラクティブエンタテインメント||Imaging apparatus, information processing system, mat, and image generation method|
|US20160327281A1 (en) *||2015-05-05||2016-11-10||June Life, Inc.||Connected food preparation system and method of use|
|US10311634B2 (en) *||2015-07-21||2019-06-04||IAM Robotics, LLC||Three dimensional scanning and data extraction systems and processes for supply chain piece automation|
|JP6329112B2 (en) *||2015-09-16||2018-05-23||東芝テック株式会社||Information processing apparatus and program|
|US10650368B2 (en) *||2016-01-15||2020-05-12||Ncr Corporation||Pick list optimization method|
|US10366379B2 (en) *||2017-01-30||2019-07-30||Ncr Corporation||Remote weigh station with delayed fraud intervention|
|US10373316B2 (en) *||2017-04-20||2019-08-06||Ford Global Technologies, Llc||Images background subtraction for dynamic lighting scenarios|
|DE102018213982A1 (en) *||2018-08-20||2020-02-20||BSH Hausgeräte GmbH||Storage system and method for operating such a storage system|
|US20190050646A1 (en) *||2018-09-07||2019-02-14||Intel Corporation||Technologies for identifying unrecognizable objects in autonomous systems|
Family Cites Families (18)
|Publication number||Priority date||Publication date||Assignee||Title|
|US3770111A (en) *||1972-05-03||1973-11-06||Fmc Corp||Apparatus for sorting fruit according to color|
|US4106628A (en) *||1976-02-20||1978-08-15||Warkentin Aaron J||Sorter for fruit and the like|
|IT1055430B (en) *||1976-02-23||1981-12-21||Tasco Spa||Method and apparatus for recognizing real-time images|
|US4515275A (en) *||1982-09-30||1985-05-07||Pennwalt Corporation||Apparatus and method for processing fruit and the like|
|US4534470A (en) *||1982-09-30||1985-08-13||Mills George A||Apparatus and method for processing fruit and the like|
|JPH0321235B2 (en) *||1982-11-09||1991-03-22||Ikegami Tsushinki Kk|
|US4574393A (en) *||1983-04-14||1986-03-04||Blackwell George F||Gray scale image processor|
|JPH0214748B2 (en) *||1983-12-08||1990-04-09||Kubota Ltd|
|JPS6155787A (en) *||1984-08-27||1986-03-20||Kubota Ltd||Fruits recognition device|
|DE3509241A1 (en) *||1985-03-14||1986-09-18||Traub Gmbh Maschinenfabrik||Method for optically determining and / or localizing an object|
|DE3545960C1 (en) *||1985-12-23||1987-07-09||Messerschmitt Boelkow Blohm||Method and arrangement for recognizing parts|
|EP0267790A3 (en) *||1986-11-12||1990-01-17||Lockwood Graders (U.K.) Limited||Method and apparatus for sorting articles|
|US5085325A (en) *||1988-03-08||1992-02-04||Simco/Ramic Corporation||Color sorting system and method|
|US5253302A (en) *||1989-02-28||1993-10-12||Robert Massen||Method and arrangement for automatic optical classification of plants|
|JPH0344268A (en) *||1989-07-12||1991-02-26||Matsushita Electric Ind Co Ltd||Background eliminating device|
|US5060290A (en) *||1989-09-05||1991-10-22||Dole Dried Fruit And Nut Company||Algorithm for gray scale analysis especially of fruit or nuts|
|US5164795A (en) *||1990-03-23||1992-11-17||Sunkist Growers, Inc.||Method and apparatus for grading fruit|
|JP3179528B2 (en) *||1991-09-04||2001-06-25||キヤノン株式会社||Color image processing apparatus and method|
- 1994-04-29 US US08/235,834 patent/US5546475A/en not_active Expired - Fee Related
- 1995-02-10 JP JP02284795A patent/JP3369772B2/en not_active Expired - Fee Related
- 1995-02-24 EP EP19950102666 patent/EP0685814B1/en not_active Expired - Lifetime
- 1995-02-24 ES ES95102666T patent/ES2151000T3/en not_active Expired - Lifetime
- 1995-02-24 AT AT95102666T patent/AT196695T/en not_active IP Right Cessation
- 1995-02-24 DE DE1995618947 patent/DE69518947T2/en not_active Expired - Fee Related
- 1995-03-10 CA CA 2144404 patent/CA2144404C/en not_active Expired - Fee Related
- 1995-03-23 AU AU15034/95A patent/AU685904B2/en not_active Expired - Fee Related
- 1995-04-10 CN CN95104385A patent/CN1092375C/en not_active IP Right Cessation
- 1995-04-28 KR KR1019950010254A patent/KR100190737B1/en not_active IP Right Cessation
Also Published As
|Publication number||Publication date|
|US10192087B2 (en)||Methods and arrangements for identifying objects|
|US9600982B2 (en)||Methods and arrangements for identifying objects|
|Krig||Interest point detector and feature descriptor survey|
|Yang et al.||Efficient illuminant estimation for color constancy using grey pixels|
|US8983200B2 (en)||Object segmentation at a self-checkout|
|Arivazhagan et al.||Fruit recognition using color and texture features|
|Zhu et al.||Learning to recognize shadows in monochromatic natural images|
|Cimpoi et al.||Deep filter banks for texture recognition and segmentation|
|De Carvalho et al.||Exposing digital image forgeries by illumination color classification|
|Zhu et al.||Multi-scale color local binary patterns for visual object classes recognition|
|US20150310601A1 (en)||Methods and arrangements for identifying objects|
|Salvador et al.||Shadow identification and classification using invariant color models|
|US9224184B2 (en)||Methods and arrangements for identifying objects|
|Serby et al.||Probabilistic object tracking using multiple features|
|Malik et al.||Contour and texture analysis for image segmentation|
|Guo et al.||Single-image shadow detection and removal using paired regions|
|US8379920B2 (en)||Real-time clothing recognition in surveillance videos|
|US20130343642A1 (en)||Machine-learnt person re-identification|
|Mirmehdi||Handbook of texture analysis|
|Bianco et al.||Color constancy using faces|
|Luo et al.||Color object detection using spatial-color joint probability functions|
|Unay et al.||Stem and calyx recognition on ‘Jonagold’apples by pattern recognition|
|Sengupta et al.||Identification and determination of the number of immature green citrus fruit in a canopy under different ambient light conditions|
|US9042650B2 (en)||Rule-based segmentation for objects with frontal view in color images|
|EP1452995B1 (en)||Method for detecting color objects in digital images|