T 1419/06 30-10-2008

I. The appellant (opponent) lodged an appeal, received on 13 September 2006, against the decision of the opposition division, dispatched on 11 August 2006, to reject the opposition against the European patent No. 1 042 645. The fee for the appeal was paid on 13 September 2006. The statement setting out the grounds of appeal was received on 15 December 2006.

II. Opposition had been filed against the patent as a whole on the basis of Article 100(a) EPC on the ground that the subject-matter of the patent was not patentable within the terms of Articles 52 to 57 EPC because it lacked novelty and did not involve an inventive step. To support its objections the opponent referred inter alia to the following documents:

(E1) VDI Berichte Nr. 1006, 1992, pages 23 - 38, K.-H. Breyer et al "Holometrische Koordinatenmeßtechnik - Neue Ansätze zur umfassenden Messung beliebiger Formelemente";

(E3) Carl Zeiss Germany, "HOLOS-UX Messen und Digitalisieren von Freiformflächen", Company's Leaflet, 1993;

(E7) IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 14, No. 2, February 1992, pages 239 - 256, P.J. Besl et al. "A Method for Registration of 3-D Shapes";

(E8) JP-A-5 010 751.

Documents E1 and E3 had been filed with the notice of opposition. Documents E7 and E8 were filed with the grounds of appeal. Furthermore the appellant (opponent) filed an auxiliary request for oral proceedings.

III. In a reply of 19 June 2007 the respondent requested that the appeal be dismissed and that the patent be maintained as granted and also filed an auxiliary request for oral proceedings. Furthermore in a subsequent letter of 31 August 2007 the respondent objected to the late-filed documents E7 and E8, submitting that these were not more relevant than the documents already on file and should therefore be disregarded.

IV. In a further letter of 8 February 2008 the appellant for the first time in the opposition appeal proceedings raised an objection under Art. 123(2) EPC against the independent claims of the patent as granted and requested that the patent be revoked.

V. In an official communication of 8 July 2008 oral proceedings were appointed to take place on 30 October 2008.

VI. In a letter filed on 26 September 2008 the respondent objected to the newly filed ground of opposition as being inadmissible and submitted it should not be considered as a fresh ground for appeal. It submitted further arguments as to the substance of the independent claims and filed five sets of claims as auxiliary requests.

VII. On 30 October 2008 oral proceedings were held.

VIII. At the oral proceedings the appellant requested that the decision under appeal be set aside and that the patent be revoked.

IX. The respondent requested that the patent be maintained as granted or alternatively that the decision under appeal be set aside and the patent be maintained on the basis of the claims of one of the auxiliary requests filed on 26 September 2008.

X. The wording of independent claim 1 of the patent as granted (main request) reads as follows (the numbering of features "a1)" to "f1)" is not part of the claims, but has been introduced for easier reference in the following Reasons):

"An apparatus (10) for finding shape deformations in objects comprising:

a1) an imaging device (12) for obtaining a scanned image of said object;

b1) a memory (30) adapted to store a reference image of said object;

c1) an image register (24) coupled to said imaging device (12) and to said memory (30) to store patch information corresponding to said reference image (30) and said scanned image (22);

d1) a transformation estimator (26) coupled to said image register (24) to provide a transform for comparing said scanned image (22) to said reference image (30);

e1) a deformation estimator (28) coupled to said transformation estimator (26) and to said image register (24) to utilize said transform and said patch information to determine shape deformations of said object (12);

characterised by

f1) said transformation estimator including a comparator arranged to compare a plurality of respective patches of selectable size and number from each of said reference image and said scanned image".

The wording of independent claim 7 according to this request reads as follows:

"A method of finding shape deformations in objects comprising:

obtaining a scanned image of said object;

storing a reference image of said object in a memory; storing patch information corresponding to said reference image and said scanned image (22) in an image register (24) coupled to an imaging device (12) and to said memory;

providing a transform for comparing said scanned image (22) to said reference image by means of a transformation estimator (26) coupled to said image register (24); and

utilizing said transform and said patch information to determine shape deformations of said object (12) by means of a deformation estimator (28) coupled to said transformation estimator (26) and to said image register (24);

characterised in that

said step of providing a transform comprises comparing a plurality of respective patches of selectable size and number from each of said reference image and said scanned image".

The wording of independent claim 20 of this request reads as follows:

"A storage medium encoded with machine-readable computer program code for performing the method of claim 7".

Claims 2 to 6 and claims 8 to 19 are dependent claims.

The claims of the auxiliary requests are not relevant for the purpose of this decision.

XI. The arguments of the appellant may be summarised as follows.

During the examination proceedings independent claims 1 and 7 have been amended to include the feature "said transformation estimator including a comparator arranged to compare a plurality of respective patches of selectable size and number from each of said reference image and said scanned image". However, nowhere in the original patent application documents is there a disclosure of a feature "comparator to compare ...patches ...from each of said reference image and said scanned image". In particular there is no explanation whatsoever what should be understood by the expression "...to compare...patches from each of said reference image and said scanned image". In the decision under appeal it had been argued that the subject-matter of the independent claims was novel by virtue of this feature f1). However, since this feature had not been disclosed in the original patent application and, furthermore, is obscure, feature f1) includes an inadmissible extension of subject-matter under Article 123(2) EPC. The patent should therefore be revoked.

The reason for submitting documents E7 and E8 with the grounds of appeal was motivated by the sudden change in the position of the opposition division during the oral proceedings compared to its prior assessment in the preliminary opinion of 19 January 2006, in which documents E1 to E3 had been considered to anticipate the subject-matter of the independent claims. This change of position appears to be related to the fact that the patent proprietor has attributed to some features in claims 1 and 7 a meaning which differs from the meaning which these features have if the patent specification is read in an objective way. In this respect Article 69, paragraph 1, second phrase EPC clearly specifies that for the interpretation of the claims the description and drawings shall be used to interpret the claims. In particular the feature "patch" has, according to paragraph [0012] of the patent specification, a very broad meaning of "...a grid on the reference plane which grid encompasses at least a portion of the surface of the imaged object". Also the expression "patch information" in feature c1) of claim 1 is vague and includes, according to paragraph [0028] of the patent specification, merely some indefinite kind of information on the subdivision of the "reference image" and the "scanned image". Furthermore, the patent proprietor has alleged, as is also suggested by the wording of claim 1, that for the alignment process no measurement data points and reference image points would be needed because patches would be compared. This allegation is in complete contradiction to the disclosure of the patent in suit. For instance, paragraph [0037] of the specification discloses that "...In one embodiment of the invention, robust registration of the data points using low curvature patches is performed by a Robust-Closest Patch algorithm (RCP) that accurately registers the data points to the CAD model, using all points on all visible and stable surfaces of the part". Therefore single data points are registered to the CAD model. Also the second algorithm described in this paragraph is based on minimising the distance between two points: "...In one embodiment of the invention, registration using RCP is driven by low curvature patches computed from the model off-line. The RCP algorithm uses an approximate normal distance between a patch and a surface..." As clearly follows from paragraphs [0044] to [0049] and corresponding Figures 3 and 4, the aforesaid distance ("approximate normal distance") is determined by a point p on a model patch and a point q of the measured image data, wherein the resulting distances are minimised according to equation (1) of the patent specification, i.e. the Gauss-criterion, which minimises the sum of the squared distances. See also paragraph [0055] according to which "since the local patch P' has no detectable boundaries or landmarks" point(s) p' cannot be determined. Instead points qi are determined by dropping a perpendicular

FORMULA/TABLE/GRAPHIC

from the model patch to the data point cloud. Therefore, in fact, these points qi are the "patches" representing the scanned image data. In its decision the opposition division unexpectedly interpreted claim 1 in a completely different way than in its earlier communication, as a result of which the opponent/appellant was forced to refer to the new documents E7 and E8 in arguing its case.

With respect to the issue of patentability, document E1 discloses the software package "Holos" which runs on a computer of a coordinate measurement machine (CMM). In this apparatus deviations of measured data points from surface areas of a reference model are determined. In the example shown in Figure 10 of E1 the reference model is defined by freeform surfaces, in E1 referred to as "patches", and in the case of Figures 11 to 16 by standard analytical elements; in the case of Figure 9 both types are used. Any deviation of the measured data points of the workpiece from the reference model is a "shape deviation" in the meaning of paragraph [0002] of the patent specification. Therefore document E1 discloses an apparatus for finding shape deformations in objects comprising:

a1) an imaging device (CCM with computer) for obtaining a scanned image of the object (measured data of the workpiece);

b1) a memory (of the computer) adapted to store a reference image of the object (reference patches of the workpiece);

c1) an image register (computer memory) coupled to the imaging device and to the memory to store patch information corresponding to the reference image and the scanned image (every measured data point is allocated to a surface area or reference patch, see Figure 10; as explained before, both in the patent as well as in E1, points qi are determined by dropping a perpendicular

FORMULA/TABLE/GRAPHIC

from the model patch to the data point cloud. Therefore patch information of the corresponding reference patch for every measured data point is stored and as a result all data points are classified as "areas" to the corresponding reference patch. These data points comprise a grid and are therefore "patches" within the extremely broad definition of paragraph [0012] of the patent);

d1) a transformation estimator coupled to the image register to provide a transform for comparing the scanned image to the reference image (the part of the program carrying out the 3D-best fit alignment);

e1) a deformation estimator coupled to the transformation estimator and to the image register to utilize the transform and the patch information to determine shape deformations of the object (the part of the program in which the deviations of the measured data fitted to the reference patches of the workpiece are determined); and

f1) wherein the transformation estimator includes a comparator arranged to compare a plurality of respective patches of selectable size and number from each of the reference image and the scanned image (fitting and alignment of the measured points allocated to the respective reference patch). With respect to the feature "selectable size and number" it is pointed out, that in feature f1), the point in time the size and number of the patches "to be compared" is scheduled, is left completely open. In the appellant's opinion this condition is also met if the size and number of the patches is variable by, for instance, a modification of the reference model of the workpiece.

It follows that the features of claim 1, insofar as these are clear and originally disclosed, by having recourse for the interpretation of the patent specification, are anticipated by the Holos system disclosed in document E1 (Art. 52(1) and 54 EPC). Since independent claim 7 defines the same technical features as claim 1 as corresponding process steps its subject-matter is equally known from document E1.

In its decision the opposition division had argued that the apparatus of claim 1 differed from the prior art Holos system disclosed in document E1 in feature f1), because in the latter system all data points of the scanned workpiece were used for fitting and aligning with the reference patch image. Therefore, instead of comparing "a plurality of patches ...from each of the reference image and the scanned image" as defined in feature f1) of claim 1, in the Holos system "a plurality of patches ...from the reference image with all measured points of the scanned image" was compared. Since, as explained before, a plurality of measurement points defines a grid and therefore a patch, the only difference between feature f1) and the prior art is that according to this feature the number of used data points can be reduced, which may be found beneficial for a faster data reduction process. Although, as already argued, the patent specification did not address a technical problem to be solved, the objective technical problem underlying this single difference over the prior art may be seen in offering a faster fitting and alignment. Clearly, this problem is obvious and the solution, to reduce the number of used data points in the case of a very large number of such points, was known in the art. This is illustrated by reference to document E7, which discloses a very similar registration process to the Holos process. This document shows several examples in which, if desired, the number of measurement data points can be considerably reduced (for instance, using planar section cuts and thinning out the data using a chord length deviation check, see page 252, Section 3 "Terrain Data"). In any case it is the appellant's firm belief that this idea, of thinning out the number of measurement points if this appears to be too large for an efficient data reduction or registration process, only forms part of the ordinary general knowledge of the skilled person in the field of CAD and CCMs and therefore does not require an inventive step. Hence the independent claims are not allowable (Art. 52(1) and 56 EPC).

XII. The arguments of the respondent may be summarised as follows.

With the submission of 8 February 2008, the appellant raised for the first time in the opposition and the opposition appeal proceedings an objection under Article 123(2) EPC with respect to claims 1 and 7 as granted. This objection, forming a fresh ground of opposition, is inadmissible without the approval of the patentee in accordance with the Decisions G 9/91 and G 10/91 of the Enlarged Board of Appeal and the respondent patentee explicitly withholds such approval. Therefore it is requested that the fresh ground of opposition now raised by the opponent be dismissed.

With respect to documents E7 and E8, in the opinion of the respondent these documents should be disregarded because they have been submitted belatedly and, moreover, are of no relevance. In particular document E7 describes a so-called iterative closest point (ICP) algorithm which has nothing to do with a comparison of patches of selectable size and number from each of the reference image and the scanned image as defined in claims 1 and 7. In particular this is illustrated by the Section VII "Conclusions" on page 253 of this publication, where it is disclosed: "if a data shape were to come in a form other than point set form, a dense set of points on the data shape can serve as the data point set". This clearly demonstrates that in document E7 only points are of interest, thus neither this document nor document E8 adds anything to the disclosures of the other documents. Hence documents E7 and E8 are not relevant for the independent claims and should consequently be disregarded.

The allegation of the appellant making reference to paragraph [0012] of the patent specification that the term "patch" would have a very broad meaning and that, if interpreted by having resort to the specification, would include individual probing points, is traversed: Claim 1 includes the following phrase: " ...to compare a plurality of patches of selectable size and number...". A single probing point does not have a selectable size, because it is a one-dimensional object and neither has a size nor an orientation. Furthermore, according to paragraph [0012] a patch is defined by "...a grid, which shall encompass a portion of the surface of the imaged object". A probing point is not a grid and, being a one-dimensional object, does not encompass a portion of the surface.

The appellant has further, by referring to paragraphs [0037] and [0043] to [0047], alleged that the patent does not disclose a comparison of patches. In paragraph [0037] it is disclosed that the registration of the data points using low curvature patches is carried out by a Robust-Closest Patch (RCP) algorithm. From this paragraph it immediately follows that:

- the term "Robust-Closest Patch algorithm" clearly indicates that patches are used;

- RCP matches model patches to data surfaces;

- RCP uses an approximate normal distance between a patch and the data surface;

- data points are used to generate patches.

The RCP algorithm is further described in paragraphs [0044] to [0049]. Paragraph [0046] includes: "Given the approximate viewpoint, the model surface can be digitized at regular grid points on an image plane, local curvature patches retained, giving a set of regularly spaced patches. Each patch Pi is represented by its centre position pi and its outward normal

FORMULA/TABLE/GRAPHIC

as shown in Figure 4". In this context "centre position" means that the position of pi is in the centre of the patch. The next paragraph states that the RCP algorithm "...translates the model patches to align the model's centre of mass with that of the data". Alignment of model to data patch is therefore carried out using centre of mass. Finally the points pi and qi representing the centres of mass are not probing points. This becomes immediately apparent from paragraph [0048], the first two sentences of which read: "In step 84, RCP algorithm 80 finds for each patch Pi the matching location qi by moving a matched filter, sized to Pi' along the line Ii' through pi and parallel to

FORMULA/TABLE/GRAPHIC

searching for the nearest significant response from current location pi' as shown in Figure 4. This estimates the piercing point of Ii' with the implicit surface from which the data are measured without the expensive and noise sensitive process of estimating surface parameters from the data". This point Ii' is the piercing point of the normal

FORMULA/TABLE/GRAPHIC

with the implicit surface of the data point, therefore the piercing point is not a probing point but a point on the implicit surface generated from the data points. Hence, on this basis it is evident that patches are compared using their centres of mass and their normals. The RCP algorithm therefore uses for the alignment patches, which are generated from the probing points and the computer model. Finally the appellant's interpretation of paragraph [0055] is disputed: points qi cannot be found by dropping a perpendicular

FORMULA/TABLE/GRAPHIC

from the model patch to a "data point cloud", since in order to obtain an intersection point a surface must be present. Therefore the features of the claims, in particular concerning "patches" and "comparing ...patches" are unambiguously supported by the description of the patent specification and their interpretation for the issue of patentability is unequivocal.

Document E1 describes two separate aspects of the Holos software. A first aspect, summarised in the first paragraph of Section 4.2 of this document, relates to the digitalization of models (for instance, prototypes), in which a computer model is generated by scanning the surface of the model to obtain a plurality of probing points and subsequently reconstructing the surface of the model by using the probing points. This can be carried out by parameterised Béziér surfaces which are determined by using the probing points as control points. Thus in the digitization process a computer model is generated for a prototype. The second aspect of the Holos software package, described in the remaining paragraphs of this Section 4.2, relates to the best-fit alignment of a computer model with probing points obtained by scanning a workpiece. These two different aspects of the Holos software are also distinguished on page 2, second and third paragraph, of document E3, from which it clearly follows that these are separate embodiments which cannot be combined for this issue of patentability. With respect to the second aspect of Holos document E1 discloses on page 32, first paragraph, that in order to align the measured data with the computer model all probing points are used, which, according to the next sentence, is indispensable ("unverzichtbar"). In Figures 11 to 16 of E1 a best-fit alignment of objects, in this case standard analytical elements, is shown. Therefore in E1 measured probing points are compared with the surface of a computer model. Thus the method used in the Holos software package differs from the apparatus of claim 1 and the method of claim 7 of the patent, because these require [in feature c1)] the storing of patch information corresponding to the reference (i.e. the computer model) image and the scanned image (of the workpiece) and in addition [in feature f1)] comparing a plurality of respective patches of selectable size and number from each of the reference image and the scanned image. Therefore the subject-matter of the independent claim is novel.

The subject-matter of the claims also involves an inventive step because all prior art documents use discrete points for comparing the scanned image with the reference image for transforming the scanned image to provide alignment of the scanned image with the reference image. Furthermore the prior art does not disclose or suggest a selection of the size and number of patches. Hence the claims also involve an inventive step.

Therefore it is requested to dismiss the appeal and to maintain the patent as granted.