Errata

it should read

'... to build and install OpenCV, update DYLD_LIBRARY_PATH ...'

instead of

'... to build and install OpenCV, update LD_LIBRARY_PATH ...'

(see 'man dyld')

The comment in the code :

// While loop (as in Example 2) capture & show video

This causes an error, the "capture" needs to be replaced with "g_capture", as this version of the code creates a global variable, and is prefaced with a "g_".

Note from the Author or Editor:
Submitter is correct. I believe that this was in a previous erratum. In any case, please make the change indicated: replace 'capture' with 'g_capture'.

In Example 2-5, third to the last line
cvPyrDown requires three parameters, in, out and filter (which is passed in the function as parameter). So this line should be
cvPyrDown( in, out, filter)

Note from the Author or Editor:
Submitter is correct. He is referring to the line immediately before the code line "return( out );". Please make the indicated change.

The code:

IplImage* out = cvCreateImage(
cvSize( cvGetSize( in ),
IPL_DEPTH_8U,
1
);

does not compile. cvGetSize takes two arguments. It can be fixed thusly:

IplImage* out = cvCreateImage(
cvGetSize( in ),
in->depth, //IPL_DEPTH_8U,
1);

This code is correct and compiles.

Change "Although the CVMat structure" to "Although the CvMat structure"

Change "(or, more likely, cvArr*)" to "(or, more likely, CvArr*)"

Change "Absolute value" to "Average value"

Function definitions for absolute value functions should be

void cvAbs(
const CvArr* src1, CvArr* dst
);

void cvAbsDiff(
const CvArr* src1, const CvArr* src2, CvArr* dst
);

void cvAbsDiffS(
const CvArr* src, CvArr* dst, CvScalar value
);

Delete code block:
"void cvConvertScale(
const CvArr* src,
CvArr* dst,
double scale = 1.0,
double shift = 0.0
);"

It's stated that cvGetSubRect(...) takes as it's second argument a CvArr* submat, where as it only takes CvMat *

Note from the Author or Editor:
Code should read:

CvMat* cvGetSubRect(
const CvArr* arr,
CvMat* submat,
CvRect rect
);

In first sentence, change "cvGetColumns()" to "cvGetCols()"

Change "Int" to "int"

Code block will look like this:

double cvInvert(
const CvArr* src,
CvArr* dst,
int method = CV_LU
);

Change last word of paragraph "cvCvtScale()" to "cvConvertScale()."

change first line of code block from "void" to "cvNot":

void(
const CvArr* src,
CvArr* dst
);

to

cvNot(
const CvArr* src,
CvArr* dst
);

Change "The argument op controls..."
to "The argument dim controls..."

The third sentence in the paragraph:"The closely related function cvSubS() does the same thing except that the constant scalar value is added to every element of src."

should be

"The closely related function cvSubS() does the same thing except that the constant scalar value is subtracted from every element of src."

Also delete the first two sentences of the paragraph, so that the paragraph starts with "The closely related function"

cvSub has been explained twice in subsection "cvSub" and "cvSub, cvSubS and cvSubRS"

Note from the Author or Editor:
Delete the heading cvSub, the code block and the paragraph following it. (the whole section for cvSub)

The above function takes two parameters as input, the second one being a CvRect struct which specifies the rectangle to draw. However, in the function itself the call to cvRectangle references box and makes no reference to rect. I think all of the references to box in this function should be references to rect.

Also replace cvScalar with CV_RGB.

Note from the Author or Editor:
The first code block should look like this:
//
void draw_box( IplImage* img, CvRect rect ) {
cvRectangle (
img,
cvPoint( rect.x, rect.y ),
cvPoint( rect.x + rect.width, rect.y + rect.height ),
CV_RGB(0xff,0x00,0x00) /* red */
);
}

"...you can pass -1 to cvCreateCameraCapture(), which will cause OpenCV to open a window that allows you to select the desired camera."

This is incorrect. As correctly stated on p26, a value of -1 means to just select the default camera.

Note from the Author or Editor:
Submitter is partially correct. Please insert the words "on some platforms" before the referenced text "...you can pass..." (thus it should read "...on some platforms, you can pass...")

In first paragraph on page no. 110 under chapter 5: Image Processing, the name of function "cvSmooth()" is printed as "cv_Smooth()" in line "The cv_Smooth() function has four parameters ....".

Change table description to:
Table 5-1. Types of smoothing operations, meaning of their parameters, and the depth and number of channels (Nc) supported by each operation.

Change titles to:
Smooth type Name In place Nc src depth dst depth Brief description

Change Nc category in first and second row to "Any" and from "1,3" in third row to "1,3,4"

In Figure 5-6 both Image A and A-morph-B are the same. Image A should be the lightly dotted interior figure in Image A-morph-B.

Change "binary to Boolean" in third sentence.
Sentence should read:

"Observe that if the image is not Boolean then the min and max operators play a less trivial role."

Change "binary" to "Boolean." Sentence should read:

"Opening is often used to count regions in a Boolean image."

Change "not just a binary mask" to "not just a Boolean mask"

The function prototypes for cvPyrDown and cvPyrUp list the third parameter, filter, as type IplFilter which as near as I can determine isn't defined. The actual prototypes in the software I have they are type 'int'.

Note from the Author or Editor:
Change "IplFilter" to "int"

Code blocks should be:
void cvPyrDown(
IplImage* src,
IplImage* dst,
int filter = IPL_GAUSSIAN_5x5
);

and

void cvPyrUp(
IplImage* src,
IplImage* dst,
int filter = IPL_GAUSSIAN_5x5
);

In table 5-5, the Operation for the threshold type CV_THRESH_TRUNC has a mistake (probably typo?), uses M instead of T, it should read:

dst_i = (src_i > T) ? T : src_i

The result line is truncated for some of the charts. It should rise to max_value after the last vertical dotted line in the "Threshold Binary" chart. It should fall to zero after the last vertical dotted line in the "Threshold Binary, Inverted" and "Threshold to Zero, Inverted" charts.

Change "binary image" to "Boolean image"

Text should read:

"Binary threshold versus adaptive binary threshold: the input image (top) was turned into a Boolean image using a global threshold (lower left ) and an adaptive threshold (lower right); raw image courtesy of Kurt Konolidge"

Change paragraph to read:

"We can, of course, express this procedure in the form of an equation. If we define the image to be I(x, y), the kernel to be G(i, j) (where 0 ≤ i ≤ Mi ?1 and 0 ≤ j ≤ Mj ?1, with Mi, Mj being the extent of the kernel in the x- and y-dimensions respectively), and the anchor point to be located at (ai, aj) in the coordinates of the kernel, then the convolution H(x, y) is defined by the following expression:"

*(In Mi, Mj, ai, aj: i and j are subscripts)

Change "aperture_sizes" to "aperture sizes" and change code style to regular font

Change "srcTri[]" to "dstTri[]".

delete the last "}"

Change Ds to Dd

Delete + from fifth line of code.
It should read:

void cvWarpPerspective(
const CvArr* src,
CvArr* dst,
const CvMat* map_matrix,
int flags = CV_INTER_LINEAR | CV_WARP_FILL_OUTLIERS,
CvScalar fillval = cvScalarAll(0)
);

Change paragraph to read:

"Output of the code in Example 6-3 and in Example 6-4 are shown in Figure 6-14 for affine and perspective transformations. Compare this with the diagrams of Figure 6-13 to see how this works with real images. In Figure 6-14, we transformed the whole image."

(add "Example 6-4" and delete the last two sentences)

Paragraph should read:
"In order to understand the last argument, nonzero_rows, we must digress for a moment to explain that in general, DFT algorithms will strongly prefer vectors of some lengths over other lengths, or arrays of some sizes over other sizes. In most DFT algorithms, the preferred sizes are powers of 2 (i.e., 2n for some integer n). In the case of the algorithm used by OpenCV, the preference is that the vector lengths, or array dimensions, be 2p3q5r, for some integers p, q, and r. Hence the usual procedure is to create a somewhat larger array and then use cvGetSubRect() to copy your array into the somewhat roomier zero-padded array. For convenience, there is a handy utility function, cvGetOptimalDFTSize(), which takes the (integer) length of your vector and returns the first equal or larger size which can be expressed in the form given (i.e. 2p3q5r)..."

This formula states that the tilt_sum() calculation is

tilt_sum(X,Y) = sum_{y <= Y} sum_{abs(x - X) <= y} image(x, y)

but this is incorrect. The proper formula, given at the willowgarage website (http://opencv.willowgarage.com/documentation/cpp/imgproc_miscellaneous_image_transformations.html) is

tilted(X,Y) = sum_{y < Y} sum_{abs(x - X + 1) <= Y - y - 1) image(x, y)

I have confirmed that the OpenCV code is actually performing the latter and can provide a small Python program demonstrating the difference.

Note also that the previous formula refers to sqsum(X, Y) as "sum(X, Y)" and that "tilted" is the correct name of the argument to cvIntegral().

change
"int dim_j_bin_j_upper_bound = histogram->thresh2[ i ][ j+1 ];"
to

"int dim_i_bin_j_upper_bound = histogram->thresh2[ i ][ j+1 ];"

change "CvHistogram()" to "CvHistogram" (delete parentheses)

There is an error in the correlation mathematical formula in the denominator:

A summation is missing, it must be sqrt(sum(H1'?)*sum(H2'?)) in the denominator (the second summation is missing in the book, or at least, in the PDF version).

cvNormalizeHist( hist[i], 1.0 ); // Normalize it

The above is incorrect, as hist is a pointer to a CvHistogram, rather than an array of histograms. Further more, there is no variable 'i' to index any array.

Note from the Author or Editor:
The submission is correct. On the line idicated the characters "hist[i]" should be changed to simply read "hist" (i.e. the "[i]" should be removed.)

The formular for the Cross Correlation shows, that the product of the template and image must be squared. But that's wrong. See: http://opencv.willowgarage.com/documentation/image_processing.html#matching

Note from the Author or Editor:
I can?t say I understand this person?s remark, but in discussion of Kouichi?s comments, it is clear that the change required is to remove the ?squared? on the right hand side.

in the CV_TM_CCOEFF template matching formula, the sum should be above the division mark and not under as it is.

Note from the Author or Editor:
I believe that this is correct. i.e. that the transcription from the ?serialized? equation (which can be found on the opencv web page) is not correct: ?1/(w.h).Sum(?)? is not intended to imply that the sum is in the denominator, but rather that the entire sum is multiplied by ?1/(w.h)? (i.e. normalized).
[Vadim verified this assertion]

Also, the first equation in this set has a ?squared? on the right hand side which should be removed (same error as CV_TM_CCORR above).

After ..."then the entire sequence is copied." at the top of the page, add the following to the end of the paragraph:

"The return value of cvCvtSeqToArray() is a void* pointer which is equal to the elements pointer provided when the function was called."

*(void and elements should be in code font)

Figures 8-4 (pg 240) and 8-13 (page 261) have an error on the freeman chain code representation. The code that ends 5-6-6-6-6-7 should be
5-6-6-6-6-6-7; 5-sixes. This can be seen by two methods. In both figures the there are 5 left-horizontal arrows. Also in Fig 8-13 in the 45-degree view there are 5-UL arrows and the chain code ends 6-7-7-7-7-7-8; 5-sevens.

Also in comparing figure 8-3 CV_RETR_LIST to the related text: The text
says "retrieves all the contours". and "eight contours" are found". This happens to agree with the figure.

However the trees for CV_RETR_EXTERNAL, CV_RETR_CCOMP and CV_RETR_TREE ,
which also say "retrieves all the contours", have 9 contours. Also the
figure 8-2 shows 9 contours. The contour h000 is missing from the
CV_RETR_LIST tree.

Change "This code references CvTermCriteria()" to "This code references CvTermCriteria"

The code that ends 5-6-6-6-6-7 should be 5-6-6-6-6-6-7; 5-sixes. The code that ends 6-7-7-7-7-8 should be 6-7-7-7-7-7-8; 5-sevens.

Change 126.7 to 125.7

Change direction of slashes.

fprintf(fptrb,"/n");fprintf(fptrg,"/n");fprintf(fptrr,"/n");

change to

fprintf(fptrb,"\n");fprintf(fptrg,"\n");fprintf(fptrr,"\n");

formula Nchannels X max(|pt2x - pt2x| + 1,|pt2y -pt2y| + 1)
should be Nchannels X max(|pt2x - pt1x| + 1,|pt2y -pt1y| + 1)

Note from the Author or Editor:
Page number of error:
270

Location on the page:
1st paragraph

Version of product where error was found:
Printed

Detailed description of error: formula Nchannels X max(|pt2x - pt2x| + 1,|pt2y -pt2y| + 1)
should be Nchannels X max(|pt2x - pt1x| + 1,|pt2y -pt1y| + 1)

In the second for loop of the function, variable 'i' is declared in the for loop which means that when you get to the end of the function "return (i)" there is no variable in the scope.

Declare "int i" immediately before the for loop.

Note from the Author or Editor:
The submittor is correct about the error, but submitted correction is in poor style. To fix the problem, the following changes should be made:

1) The line "int found_codeword = -1;" on p282 should be inserted before the line "for(int i=0; i<c.numEntries; i++) {"

2) the line "found_codeword = i;" should be inserted immediately before the line "break;" at the bottom of page 282.

3) The line "if( i == c.numEntries ) //if no existing codeword found, make one" on p283 should be changed to read "if( found_codeword==-1 ) // If no existing codeword was found, make one."

4) The line "return(i)" on p284 should be changed to read: "return( found_codeword >= 0 ? found_codeword : c.numEntries );"

In the for loop of the function, variable 'i' is declared in the for loop which means that when you get to the end of the function "if(i >= ...)" there is no variable in the scope.

Declare "int i" immediately before the for loop.

Note from the Author or Editor:
The submittor is correct, this is a mistake. The correction proposed by the submittor would work, but it is poor coding style. The following changes should be made to make the code work correctly:

1) The line which reads: "break; //Found an entry that matched all channels" should be changed to read "return(0); // Found an entry that matched all channels."

2) The line which reads "If( i>= c.numEntries ) return(255);" should be deleted

3) The line which reads "return(0);" (immediately following the above mentioned line) should be changed to read "return 255; // Apparently no match could be found."

Switch (255) and (0) so the last two lines of code read:

if(i >= c.numEntries) return(0);
return(255);
}

I believe that this diagram is slightly flawed. In (c) Two of the "edges" are still visualized (the ones headded to the fictitious point in the south), and the "circum-circle" mentioned in the preceeding text (the one that touches that fictitious point and the two points from which the above mentioned lines are emenating) is both not in the right place and shown as a dotted line instead of a dashed line as are all of the other circles. [If the dotted edges are headed for that far away point, the visualized circle's location and included points are not consistant with the location of the far away point.]
[In conversation with Kouichi-san, Vadim suggested just getting rid of this construction in both the image and the (last sentence of the) preceeding text. I am inclined to agree.

Change the uppercase I to lower case i in the last line of both code blocks:

They should read on 309:

(CvSubdiv2DPoint*)cvGetSeqElem( (CvSeq*)subdiv, i );
}

and on 310:

CvQuadEdge2D*)cvGetSeqElem( (CvSeq*)(my_subdiv->edges), i );
}

Change:

1.Until you have circumnavigated the convex hull, go to the next edge on the hull via cvSubdiv2DRotateEdge(CvSubdiv2DEdge edge, 0 ).

to:

1.Until you have circumnavigated the convex hull, go to the next edge on the hull via cvSubdiv2DRotateEdge(CvSubdiv2DEdge edge, 2 ).

Correct "v_3" on right side of upper right figure to be v_4. Correct {v_3} in subcaption to the same figure to be {v_i}.

The numerator on the right side should be sigma-1-hat-squared, not sigma-1-squared.

Left hand side should be x-1-hat, not x-1

Change:

CvMat* error_cov_pre; // prior error covariance:
// (P_k'= F P_k-1 Ft) + Q

to:

CvMat* error_cov_pre; // prior error covariance:
// P_k'=(F P_k-1 Ft) + Q

(move the first parenthesis over)

The author says:

This image sucks. Redraw with ?real? Gaussians.

Delete ) from beginning of line.
Line should read:

typedef struct CvConDensation

Should CD->flConfidence[i] = Prob;

read

Should CD->flConfidence[i] = p;

Change "A point Q = (X, Y, Z) is projected onto the image plane by the ray passing through the center of projection, and the resulting point on the image is q = (z,y,f); the image is really just the projection screen "pushed" in front of the pinhole (the math is equivalent but simpler this way)

to:

A point Q = (X, Y, Z) is projected onto the image plane by the ray passing through the center of projection; the image plane is really just the projection screen ?pushed? in front of the pinhole (the math is equivalent but simpler this way)

Change subscript scr to src

Tangential distortion equations should be

x_{corrected} = x + [2p_1xy + p_2(r^2+2x^2)]
y_{corrected} = y + [p_1(r^2+2y^2) + 2p_2xy]

The text is:

"The homography matrix H that maps a planar object?s points onto the imager is then described completely by H = sM[r1 r2 t], where: q = sHQ'"

Strictly interpreted, though, the s term is squared that way. It might be worthwhile to reiterate that s is being factored out of H as per convention.

Note from the Author or Editor:
The submittor is correct that this is not as clear as it could be. I propose the following changes to add clarity:

1) on p385, the equation "q~ = s H Q~" should be changed by the addition of a subscript '1' on 's' (i.e. 's' becomes 's_1', where '_' denotes subscripting).

2) The text "Here we have introduced the parameter s" should be changed such that 's' is again subscripted 's_1'.

3) on p386 the equation beginning "q~ = s M W Q~, where M = ..." should be changed such that 's' is subscripted 's_1'., and the same change should be made to 's' as it appears (both times) in the matrix equation "[[x][y][z]] = s M [ r_1 r_2 r_3 t ][[X][Y][0][1]] = s..."

4) the text "...described completely by H = s M [r_1 r_2 t],..." should be changed such that the 's' is subscripted by '2' (i.e. 's' becomes 's_2').

5) in the same sentence described in 4), immediately after the comma in ", where:" (i.e. before the word 'where') the following insertion should be made: "for some other arbitrary constant s_2, ".

6) The equation on the following line should be changed such that 's' is replaced by 's_1'. (i.e. 's' subscripted by '1')

7) In the next line after the equation following the sentence in 4) and 5), after the sentence "Observe that H is now a 3-by-3 matrix. please add a footnote. The footnote should read "The constants s_1 and s_2 are not necessarily equal, but what matters to us here is that they are both equally arbitrary numbers."

The final B column vector in this equation does not need the 'T' (transpose).

Typing error, the equation should look like this:

x_p = f_x * X_w / Z_w + c_x
y_p = f_y * Y_w / Z_w + c_y

In the formula, it has y_{p} = f_{y} * X^{W}/Z^{W} + c_{y}.
It should have been y_{p} = f_{y} * Y^{W}/Z^{W} + c_{y}

Note from the Author or Editor:
Submitter is correct, please make the suggested change. (Note his use of latex formatting, the point is to change the 'X' to a 'Y')

There is an error in this figure. The right hand plane should have coordinates x_r,y_r (not x_l,y_l) as printed.

...with p_l and p_r, which are the location of the point P in the coordinate frames of the two cameras.

->

...with P_l and P_r, which are the location of the point P in the coordinate frames of the two cameras.

(l, r are subscripts)

In the second and third lines, the description of the RANSAC algorithm is really LMedS and vice versa.

Should read:
"but the basic idea of LMedS is to solve the problem many times using a random subset of the points and then take the particular solution closest to the average or the median solution. RANSAC takes a subset of points..."

In the second paragraph that begins with "Correspondence is computed..." there is mention of the parameter subPixelDisparities near the end of the paragraph however this parameter can not be found in the OpenCV source. Thus there are only 16 subdisparities per pixel and it is currently not configurable.

Note from the Author or Editor:
Delete last five lines of paragraph, from "Disparities have discrete, subpixel rsolution..." to "Remember that large disparities represent closer distances."

Add the following lines to the end of the paragraph:

"Disparities have a discrete subpixel resolution which is equal to four bits of resolution below the individual pixel level. When the output image is a 32-bit floating point image, non-integer disparities will be returned. When the output image is a 16-bit integer, the disparity will be returned in 4-bit fixed-point form (i.e. multiplied by 16 and rounded to an integer)."

Change "Finally, if we rectified the images then we initialize the block-matching state (internal allocations and parameters) using cvCreateBMState(). We can then compute the disparity maps by using cvFindStereoCorrespondenceBM()."

to:

"Finally, if we have rectified the images, we can then compute the disparity maps by using cvFindStereoCorrespondenceBM(). (The block-matching state CvStereoBMState, which is passed to cvFindStereoCorrespondenceBM(), should be created in advance using cvCreateStereoBMState().)

Change "(0 for Hartley // or 1 for Bouguet stereo methods)."

to:

(0 for Bouguet
// or 1 or 2 for Hartley stereo methods).

I believe the following line is wrong:

cvPoint3D32f(i*squareSize, j*squareSize, 0);

shouldn't the x coordinate (j) be first?

cvPoint3D32f(j*squareSize, i*squareSize, 0);

Change: "...which are in CVCORE, and face detection, which is in CV."

to:

"...which are in CXCORE, and face detection, which is in CV."

Text should read:

CV_COVAR_NORMAL
Do the regular type of covariance calculation as in the previously displayed equation. If the flag is not set, the following matrix is computed instead:
&#931; &#771;=E[(X-&#956;_x )^T (X-&#956;_x )]


When each vector size is significantly larger than the number of vectors, &#931; &#771; will be much smaller than &#931; (and thus much faster to process). On the other hand, it is easy to prove that the non-zero eigenvalues of &#931; and &#931; &#771; are the same. It also follows that for any eigenvector v of &#931;, corresponding to some eigenvalue &#955;, can be easily computed from its counterpart v &#771; of &#931; &#771; with the same eigenvalue by the relation: v=(X-&#956;_x ) ( v) &#771;.

Change "...that may be missing then we can set use_surrogates to CvDTreeParams::use_surrogates..."

to:

"...that may be missing then we can set use_surrogates to true..."
(true is in code font)

Code in Example 13-4.
cvRectangle(img,cvPoint(r.x,r.y),cvPoint(r.x+r.width,r.y+r.height),colors[i%8])
should be corrected as
cvRectangle(small_img,cvPoint(r->x, r->y), cvPoint(r->x+r->width,r->y+r->height), colors[i%8]);

Note from the Author or Editor:
Submitter is correct, please make the change indicated.