S. K. Nayar and R. M Rolle, "Reflectance Based Object Recognition", International Journal of Computer Vision, 17(3):219--240, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... the accidental imaging conditions (e.g. illumination, shading, highlights, and viewpoint) color histograms are computed from color invariants [2] 3] 12] For example, simple and effective illumination independent color ratio s have been proposedby Funt and Finlayson [2] and Nayar and Bolle [7]. Further, for the dichromatic reflection model, Gevers and Smeulders [3] proved that normalized color rgb (c 1 c 2 c 3 ) is to a large extent invariant to a change in camera viewpoint, object pose, and for the direction and intensity of the incident light. In addition, the hue color H (l 1 l 2 l ....

S. K. Nayar, and R. M. Bolle, Reflectance Based Object Recognition, International Journal of Computer Vision, Vol. 17, No. 3, pp. 219-240, 1996

....that objects can be recognized by using colour information alone. Combing colour cues with colour constancy[11, 26, 7, 4] generated even more powerful colour guided object recognition systems. Colour is only one reflectance clue that researchers have used in analyzing images. Nayar and Bolle[21], Slater and Healey[24, 25] Lin and Lee[19] and Jacobs et al. 17] among others, concentrated their object identification techniques on identifying reflectance based object properties that are invariant to illumination. In texture recognition, Healey and Wang[12] developed an illumination ....

....An additional advantage of our technique over the more traditional band ratios is that spectral derivatives are used on a per pixel basis. They do not depend on neighbouring regions, an assumption that is common in other photometric methods, which use logarithms and or narrow band filters[7, 21]. The collection of spectral derivatives evaluated at different wavelengths forms a spectral gradient. This gradient is normalized albedo descriptor, invariant to scene geometry and incident illumination for smooth diffuse surfaces. Experiments on surfaces of different colours and materials ....

Nayar, S. K. and Bolle, R. "Reflectance Based Object Recognition," International Journal of Computer Vision, Vol. 17, No. 3, March 1996, pp. 219-240.

....restricted classes of shapes. Moment based methods are somewhat related to ours, in that they compute a description of image regions to match to model volumes. These methods might align regions based on their center of mass, or on higher order moments. Examples of this approach can be found in [17, 23, 37, 36, 40, 43, 44, 48]. These approaches do not extend to the recognition of a 3 D object from a single 2 D image, however. First of all, volumes of 3 D points always produce self occlusion, since different subsets of the surface of the volume are visible from different viewpoints. Therefore, the center of mass of the ....

....candidate image groups. One might use intensity based segmentation or a system that finds salient convex sets of edges ( 28] We then consider matches between image groups and model parts, with a search that may be directed with the addition of cues such as color, as was done by Nayar and Bolle[37]. Pose is determined using our current work, and then a hypothetical projection of the model may be confirmed or rejected using additional cues. The steps of this process are illustrated in numerous experiments described in [27] That system robustly matched convex object parts, but used a ....

[Article contains additional citation context not shown here]

Nayar, S. and Bolle, R., forthcoming, "Reflectance Based Object Recognition," Int. J. of Comp. Vis.

.... histograms of colour ratios computed locally from pairs of neighbouring pixels [5] across detected edges and have been used [6] In order to avoid the limitations of the global nature of the histogram representation invariant ratios were computed across the boundaries of segmented image regions [16, 15]. However, reliable image segmentation for image retrieval has proven difficult to obtain [18, 15] As an alternative to segmentation, the image is covered by compact regions where local colour features are computed. In our previous work [13] the multimodal neighbourhood signature (MNS) has been ....

S. Nayar and R. Bolle. Reflectance Based Object Recognition. International Journal of Computer Vision, 17(3):219--240, 1996.

....pixel [6] or across detected edges [7] have been proposed. However, both methods are limited due to the global nature of the histogram representation. Localised invariant features have been extracted from nearby pixels across boundaries of segmented regions for object recognition and retrieval [17, 16, 21, 14]. However, reliable image segmentation is arguably a notoriously difficult task [20, 16] As an alternative to segmentation, the image is covered by compact regions from where local colour features are computed. For example, in the FOCUS system [3] a graph of the modes of the local colour density ....

S. Nayar and R. Bolle. Reflectance Based Object Recognition. International Journal of Computer Vision, 17(3):219--240, 1996.

.... histograms of colour ratios computed locally from pairs of neighbouring pixels [5] and across detected edges have been used [6] In order to avoid the limitations of the global nature of the histogram representation invariant ratios were computed across the boundaries of segmented image regions [16, 15]. However, reliable image segmentation for image retrieval has proven difficult to obtain [18, 15] As an alternative to segmentation, the image is covered by compact regions where local colour features are computed. In our previous work [13] the multimodal neighbourhood signature (MNS) has been ....

S. Nayar and R. Bolle. Reflectance Based Object Recognition. International Journal of Computer Vision, 17(3):219--240, 1996.

....pixel [6] or across detected edges [7] have been proposed. However, both methods are limited due to the global nature of the histogram representation. Localised invariant features have been extracted from nearby pixels across boundaries of segmented regions for object recognition and retrieval [17, 16, 21, 14]. However, reliable image segmentation is arguably a notoriously difficult task [20, 16] As an alternative to segmentation, the image is covered by compact regions from where local colour features are computed. For example, in the FOCUS system [3] a graph of the modes of the local colour density ....

S. Nayar and R. Bolle. Reflectance Based Object Recognition. International Journal of Computer Vision, 17(3):219--240, 1996.

....image pixel [7] or across detected edges [8] have been used. However, both methods are limited due to the global nature of histogram representation. In the same spirit, invariant ratio features have been extracted from nearby pixels across boundaries of segmented regions for object recognition [16, 15]. Similarly, absolute colour features have been extracted from segmented regions in [19, 13] However, reliable image segmentation is arguably a notoriously difficult task [18, 15] Other approaches, split the image into regions where local colour features are computed. The FOCUS system [3] ....

S. Nayar and R. Bolle. Reflectance Based Object Recognition. International Journal of Computer Vision, 17(3):219-- 240, 1996.

....restricted classes of shapes. Moment based methods are somewhat related to ours, in that they compute a description of image regions to match to model volumes. These methods might align regions based on their center of mass, or on higher order moments. Examples of this approach can be found in [17, 23, 37, 36, 40, 43, 44, 49]. These methods do not extend to the recognition of a 3 D object from a single 2 D image, however. First of all, volumes of 3 D points always produce self occlusion, since different subsets of the surface of the volume are visible from different viewpoints. Therefore, the center of mass of the ....

....candidate image groups. One might use intensity based segmentation or a system that finds salient convex sets of edges ( 28] We then consider matches between image groups and model parts, with a search that may be directed with the addition of cues such as color, as was done by Nayar and Bolle [37]. Pose is determined using our current work, and then a hypothetical projection of the model 3 may be confirmed or rejected using additional cues. The steps of this process are illustrated in numerous experiments described in [29] That system robustly matched convex object parts, but used a ....

[Article contains additional citation context not shown here]

S. Nayar and R. Bolle, 1996, "Reflectance Based Object Recognition," International Journal of Computer Vision, 17(3): 219--240. 33

....reflectance ratio is a measure of the difference in transfer function between two pixels that is invariant to illumination and shape so long as the latter two elements are similar. Nayar and Bolle developed the concept and have shown it to be effective for both segmentation and object recognition [41]. The reflectance ratio was originally defined for intensity images and measures the ratio in albedo between two points. The albedo of a point is the percentage of light reflected by the surface. The principle underlying the reflectance ratio is that two nearby points in an image are likely to be ....

.... of two nearby pixels in an image, then we obtain the ratio of albedos, or the reflectance ratio p as shown in (2) 2) Nayar and Bolle go on to show that multiple light sources do not affect the reflectance ratio so long as the assumption of similar geometries and illumination environments holds [41]. Unfortunately, p is unbounded and ranges from 0 to infinity. A well behaved version of the reflectance ratio is the difference in intensities divided by their sum, as in (3) 41] 3) Note that the geometry and sensor dependent terms still cancel out. Unlike the ratio in (2) however, this ....

[Article contains additional citation context not shown here]

S. K. Nayar and R. M. Bolle, "Reflectance Based Object Recognition," to appear in the International Journal of Computer Vision, 1995.

....to correctly identify objects by using color cues only. Healey and Slater[11, 27] Funt and Finlayson[9] and Finlayson[6] made object recognition by color even more widely applicable, by cancelling variations in color appearance caused by illumination changes. In object recognition Nayar and Bolle[22], Slater and Healey[25, 26] Lin and Lee[18] and Jacobs et al. 16] among others concentrated in identifying reflectance based object properties that are invariant to illumination. Another school of thought in object recognition, influenced by the work of Turk and Pentland [30] developed ....

....a considerable body of work on color assumes that the incident illumination has two or three degrees of freedom. However, Slater and Healey[28] showed that for outdoor scenes, the illumination functions have seven degrees of freedom. On the other hand, greyscale object recognition methods[22, 16] take advantage of the invariance in the color distribution on an object and can not handle very well non textured scenes. As for appearance based approaches, as Mundy et al. 20] pointed out, they do not render themselves in generalizations of identifying objects or materials that should logically ....

[Article contains additional citation context not shown here]

Nayar, S. K. and Bolle, R. "Reflectance Based Object Recognition," International Journal of Com- 13 puter Vision, Vol. 17, No. 3, March 1996, pp. 219-240.

....the paradox by using colour information in order to quickly recognize potential beacons in an image, and thereafter using the geometric pattern of the the beacons to reject false recognitions and to determine the orientation of the beacon uniquely. It is worth reiterating that colour ratios [4] are invariant to the camera parameters both extrinsic and intrinsic as well as to the lighting conditions. 70mm 74mm 297mm 210mm Fig. 1. One of the two possible beacons (the other has the grey and black blocks interchanged) The dimensions shown are for standard A4 paper which allows ....

....is able to navigate unaided. 4 Conclusion There are two keys ideas in the beacon system (i) that advanced vision systems can now be created using standard components, and (ii) that robot localization requires complex, asymmetrical beacons. The recognition software was assembled using Nayar s [4] colour ratios and standard optimization software [5] Camera localization and calibration are performed by a public domain version of Tsai s, now classic, algorithm [10, 11] The design of the beacons is a most important innovation: they are asymmetrical, so allowing camera localization without ....

S. K. Nayar and R. M. Bolle. Reflectance based object recognition. International Journal of Computer Vision, 1996.

....the paradox by using colour information in order to quickly recognise potential beacons in an image, and thereafter using the geometric pattern of the the beacons to reject false recognitions and to determine the orientation of the beacon uniquely. It is worth reiterating that colour ratios [5] (x 3) are invariant to the camera parameters both extrinsic and intrinsic as well as to the lighting conditions. 2.1 Geometric properties of the beacons The two goals of the pattern painted on the beacons are to localise the beacon accurately, and to uniquely determine the beacon s ....

....function (BRDF) however we shall consider a much simpler case. If we assume that the surface is lambertian then, by only considering closely spaced points, we have the image intensity i = ael, where ae is the surface s reflectance and l is the (unknown and variable) effective illumination level [1, 2, 5, 7]. We may eliminate l by taking the ratio of the image intensities of two closely spaced points so yielding the invariant ae 1 =ae 2 which only depends on properties of the surface and is independent of the lighting. To summarise: if the intensities at nearby pixels u and v are (r u ; g u ; b u ) ....

[Article contains additional citation context not shown here]

S. K. Nayar and R. M. Bolle. Reflectance based object recognition. International Journal of Computer Vision, 1996.

....descriptor and color matching has been a central focus of color science and engineering, only recently have a number of researchers begun to explore the use of color distributions as signatures for object recognition. Their work has demonstrated that color can potentially be a strong cue [32] 26] [29] [14] 16] 30] For image retrieval from an image database, color is one of the most effective cues [8] 1] The early approaches for recognizing objects based on their color distributions were developed by Nagao et al. and by Swain and Ballard [26] 32] Although this initial work does not ....

....to deal with scenes with variations in illumination color. By histogramming ratios of RGB values of adjacent pixels, they achieve a degree of color constancy that is relatively insensitive to illumination changes. Independently, Nayar and Bolle also used reflectance ratios for object recognition [29]. Healey et al. presented various methods mostly based on the affine property of color distribution change. By assuming a finite dimensional linear surface reflectance model, changes in illumination color result in a linear transformation of global color pixel distributions in RGB space [16] 30] ....

[Article contains additional citation context not shown here]

S.K. Nayar and R.M. Bolle. Reflectance based object recognition. IJCV, 17, 1996.

....matching curves individually is to first carry out a monocular grouping, and then match the groups. This is the approach adopted by Chung and Nevatia [5] and continued in work at USC by Havaldar and Medioni [10] In this paper we match curves individually without first grouping. Nayar and Bolle [15] developed a photometric matching constraint based on the ratio of intensities across region boundaries. This ratio of intensities has good invariance to lighting conditions. However, it is less general than the neighbourhood intensity correlation presented in this paper because it only applies to ....

S.K. Nayar and R.M. Bolle, Reflectance Based Object Recognition. International Journal of Computer Vision, 17(3):219--240, 1996.

.... models has lead to the development of an efficient algorithm for computing reflectance from a single brightness image [Nayar and Bolle 1993a] Nayar and Bolle 1993b] This result was later used to develop algorithms that recognize three dimensional objects based on their reflectance properties [Nayar and Bolle 1994] . Traditionally, recognition and pose estimation algorithms have relied mainly on geometric models. In contrast, our results demonstrate the benefits of using physical attributes like reflectance, in addition to geometric constraints, for recognition. In the previous overview, we reported the ....

....mainly on geometric models. In contrast, our results demonstrate the benefits of using physical attributes like reflectance, in addition to geometric constraints, for recognition. In the previous overview, we reported the development of a new comprehensive model for diffuse reflectance [Oren and Nayar 1994b] Oren and Nayar1994c ] This model was demonstrated to be an extensive generalization of the popular Lambert s model [Lambert 1760] and has far reaching implications for machine vision, visual psychophysics, computer graphics, and remote sensing. It shows that diffuse reflection can ....

[Article contains additional citation context not shown here]

S. K. Nayar and R. M. Bolle. Reflectance based object recognition. International Journal of Computer Vision, 1994. Accepted.

No context found.

S. K. Nayar and R. M Rolle, "Reflectance Based Object Recognition", International Journal of Computer Vision, 17(3):219--240, 1996.

No context found.

S. K. Nayar and R. Bolle, "Reflectance based object recognition", Inter. J. Comp. Vision vol. 17, no.3, pp.219-240,1996

No context found.

S. K. Nayar and R. Bolle, "Reflectance based object recognition", Inter. J. Comp. Vision vol. 17, no.3, pp.219-240,1996

No context found.

S. K. Nayar and R. M Rolle, "Reflectance Based Object Recognition", International Journal of Computer Vision, 17(3):219--240, 1996.

No context found.

S. K. Nayar, and R. M. Bolle, Reflectance Based Object Recognition, Int'l Journal of Computer Vision, Vol. 17, No. 3, pp. 219-240, 1996.

No context found.

S. K. Nayar, and R. M. Bolle, Reflectance Based Object Recognition, International Journal of Computer Vision, Vol. 17, No. 3, pp. 219-240, 1996

No context found.

S. K. Nayar and R. M. Bolle, "Reflectance Based Object Recognition," International Journal of Computer Vision, 1996.

No context found.

S.K. Nayar and R.M. Bolle, "Reflectance Based Object Recognition," Int'l J. Computer Vision, vol. 17, no. 3, pp. 219-240, 1996.

No context found.

S. K. Nayar, and R. M. Bolle, Reflectance Based Object Recognition, International Journal of Computer Vision, Vol. 17, No. 3, pp. 219-240, 1996

No context found.

S. K. Nayar and R. M. Bolle, Reflectance based object recognition, Int. J. Comput. Vision 1996.

No context found.

S. K. Nayar and R. M. Bolle, "Reflectance Based Object Recognition," to appear in the International Journal of Computer Vision, 1995.

No context found.

S.K. Nayar and R.M. Bolle. Reflectance based object recognition. International Journal of Computer Vision, 17(3):219--240, 1996. 47

No context found.

Nayar SK, Bolle RM (1996) Reflectance Based Object Recognition. Int J Comput Vision 17(3): 219--240

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC