How to open DBase files (.DBF) in Matlab?

Question

I've googled and searched through Matlab Central, but cannot find any way to open DBF files directly in Matlab. There are some references to DBFREAD function in TMW File Exchange, but it's not available anymore. Is it really a problem?

I do have Database toolbox, but could not find dbf support there.

I don't want to use Excel or other tools for converting files outside of Matlab, since I have a lot of files to process. ODBC also is not good, I need the code to work under Mac and Unix as well.

Please help.

Answer 1

If you are interested only in numerical values, try xlsread command.

Answer 2

The way I see it, you have two options:

Method 1: use ODBC to read the dBASE files:

This requires the database toolbox

cd 'path/to/dbf/files/'
conn = database('dBASE Files', '', '');
cur = exec(conn, 'select * from table');
res = fetch(cur);
res.Data
close(conn)

'dBASE Files' is an ODBC Data Source Name (DSN) (I believe its installed by default with MS Office). It uses the current directory to look for .dbf files.

Or maybe you can use a DSN-less connection string with something like:

driver = 'sun.jdbc.odbc.JdbcOdbcDriver';
url = 'jdbc:odbc:DRIVER={Microsoft dBase Driver (*.dbf)};DBQ=x:\path;DefaultDir=x:\path';
conn = database('DB', '', '', driver, url);
...

if this gives you trouble, try using the FoxPro ODBC Driver instead..

For Linux/Unix, the same thing could be done. A quick search revealed these:

• How do set up an ODBC DSN on Mac or Linux/Unix
• how to create and use dBase-format files with OpenOffice

Method 2: read/write .dbf files directly with a library

There's a Java library available for free from SVConsulting, that allows you to read/write DBF files: JDBF. (UPDATE: link seems to be dead, use Wayback Machine to access it)

You can use Java classes directly from MATLAB. Refer to the documentation to see how.

Answer 3

I contacted with Brian Madsen, the author of DBFREAD function, which was deleted from File Exchange, probably because The Mathworks is going to include this function into MATLAB in some future release. Brian kindly gave me permission to publish this function here. All copyright information left intact. I only modified lines 33-38 to allow DBFREAD to read file outside working directory.

function [dbfData, dbfFieldNames] = dbfread(filename, records2read, requestedFieldNames)
%DBFREAD Read the specified records and fields from a DBF file.
%
%   [DATA, NAMES] = DBFREAD(FILE) reads numeric, float, character and date
%   data and field names from a DBF file, FILE.
%
%   [DATA, NAMES] = DBFREAD(FILE, RECORDS2READ) reads only the record
%   numbers specified in RECORDS2READ, a scalar or vector.
%
%   [DATA, NAMES] = DBFREAD(FILE, RECORDS2READ, REQUESTEDFIELDNAMES) reads
%   the data from the fields, REQUESTEDFIELDNAMES, for the specified
%   records. REQUESTEDFIELDNAMES must be a cell array. The fields in the
%   output will follow the order given in REQUESTEDFIELDNAMES.
%
%   Examples:
%
%       % Get all records and a list of the field names from a DBF file.
%       [DATA,NAMES] = dbfread('c:\matlab\work\mydbf')
%
%       % Get data from records 3:5 and 10 from a DBF file.
%       DATA = dbfread('c:\matlab\work\mydbf',[3:5,10])
%
%       % Get data from records 1:10 for three of the fields in a DBF file.
%       DATA = dbfread('c:\matlab\work\mydbf',1:10,{'FIELD1' 'FIELD3' 'FIELD5'})
%
%   See also XLSREAD, DLMREAD, DLMWRITE, LOAD, FILEFORMATS, TEXTSCAN.

% Copyright 2008 The MathWorks, Inc.
% $Revision: 1.0 $   $Date: 2008/04/18 05:58:17 $

[pathstr,name,ext] = fileparts(filename);

dbfFileId = fopen(filename,'r','ieee-le');
if (dbfFileId == -1)
    dbfFileId = fopen(fullfile(pathstr, [name '.dbf']),'r','ieee-le');
end
if (dbfFileId == -1)
    dbfFileId = fopen(fullfile(pathstr, [name '.DBF']),'r','ieee-le');
end

if (dbfFileId == -1)
    eid = sprintf('MATLAB:%s:missingDBF', mfilename);
    msg = sprintf('Failed to open file %s.dbf or file %s.DBF.',...
            name, name);
    error(eid,'%s',msg)
end

info = dbfinfo(dbfFileId);
if ~exist('requestedFieldNames','var')
    dbfFieldNames = {info.FieldInfo.Name};
    requestedFieldNames = dbfFieldNames;
else
    dbfFieldNames = (info.FieldInfo(matchFieldNames(info,requestedFieldNames)).Name);
end
fields2read = matchFieldNames(info,requestedFieldNames);

% The first byte in each record is a deletion indicator
lengthOfDeletionIndicator = 1;

if ~exist('records2read','var')
    records2read = (1:info.NumRecords);
elseif max(records2read) > info.NumRecords
    eid = sprintf('MATLAB:%s:invalidRecordNumber', mfilename);
    msg = sprintf('Record number %d does not exist, please select from the range 1:%d.',...
        max(records2read), info.NumRecords);
    error(eid,'%s',msg)
end

% Loop over the requested fields, reading in the data
dbfData = cell(numel(records2read),numel(fields2read));
for k = 1:numel(fields2read),
    n = fields2read(k);
    fieldOffset = info.HeaderLength ...
                  + sum([info.FieldInfo(1:(n-1)).Length]) ...
                  + lengthOfDeletionIndicator;
    fseek(dbfFileId,fieldOffset,'bof');
    formatString = sprintf('%d*uint8=>char',info.FieldInfo(n).Length);
    skip = info.RecordLength - info.FieldInfo(n).Length;
    data = fread(dbfFileId,[info.FieldInfo(n).Length info.NumRecords],formatString,skip);
    dbfData(:,k) = feval(info.FieldInfo(n).ConvFunc,(data(:,records2read)'));
%     dbfData(:,k) = info.FieldInfo(n).ConvFunc(data(:,records2read)');
end

fclose(dbfFileId);

%--------------------------------------------------------------------------
function fields2read = matchFieldNames(info, requestedFieldNames)
% Determine which fields to read.

allFieldNames = {info.FieldInfo.Name};
if isempty(requestedFieldNames)
    if ~iscell(requestedFieldNames)
        % Default case: User omitted the parameter, return all fields.
        fields2read = 1:info.NumFields;
    else
        % User supplied '{}', skip all fields.
        fields2read = [];
    end
else
    % Match up field names to see which to return.
    fields2read = [];
    for k = 1:numel(requestedFieldNames)
        index = strmatch(requestedFieldNames{k},allFieldNames,'exact');
        if isempty(index)
            wid = sprintf('MATLAB:%s:nonexistentDBFName',mfilename);
            wrn = sprintf('DBF name ''%s'' %s\n%s',requestedFieldNames{k},...
                     'doesn''t match an existing DBF name.',...
                     '         It will be ignored.');
            warning(wid,wrn)
        end
        for l = 1:numel(index)
            % Take them all in case of duplicate names.
            fields2read(end+1) = index(l);
        end
    end
end

%--------------------------------------------------------------------------

function info = dbfinfo(fid)
%DBFINFO Read header information from DBF file.
%   FID File identifier for an open DBF file.
%   INFO is a structure with the following fields:
%      Filename       Char array containing the name of the file that was read
%      DBFVersion     Number specifying the file format version
%      FileModDate    A string containing the modification date of the file
%      NumRecords     A number specifying the number of records in the table
%      NumFields      A number specifying the number of fields in the table
%      FieldInfo      A 1-by-numFields structure array with fields:
%         Name        A string containing the field name 
%         Type        A string containing the field type 
%         ConvFunc    A function handle to convert from DBF to MATLAB type
%         Length      A number of bytes in the field
%      HeaderLength   A number specifying length of the file header in bytes
%      RecordLength   A number specifying length of each record in bytes

%   Copyright 1996-2005 The MathWorks, Inc.
%   $Revision: 1.1.10.4 $  $Date: 2005/11/15 01:07:13 $

[version, date, numRecords, headerLength, recordLength] = readFileInfo(fid);
fieldInfo = getFieldInfo(fid);

info.Filename     = fopen(fid);
info.DBFVersion   = version;
info.FileModDate  = date;
info.NumRecords   = numRecords;
info.NumFields    = length(fieldInfo);
info.FieldInfo    = fieldInfo;
info.HeaderLength = headerLength;
info.RecordLength = recordLength;

%----------------------------------------------------------------------------
function [version, date, numRecords, headerLength, recordLength] = readFileInfo(fid)
% Read from File Header.

fseek(fid,0,'bof');

version = fread(fid,1,'uint8');

year  = fread(fid,1,'uint8') + 1900;
month = fread(fid,1,'uint8');
day   = fread(fid,1,'uint8');

dateVector = datevec(sprintf('%d/%d/%d',month,day,year));
dateForm = 1;% dd-mmm-yyyy
date = datestr(dateVector,dateForm);

numRecords   = fread(fid,1,'uint32');
headerLength = fread(fid,1,'uint16');
recordLength = fread(fid,1,'uint16');

%----------------------------------------------------------------------------
function fieldInfo = getFieldInfo(fid)
% Form FieldInfo by reading Field Descriptor Array.
%
% FieldInfo is a 1-by-numFields structure array with the following fields:
%       Name      A string containing the field name 
%       Type      A string containing the field type 
%       ConvFunc  A function handle to convert from DBF to MATLAB type
%       Length    A number equal to the length of the field in bytes

lengthOfLeadingBlock    = 32;
lengthOfDescriptorBlock = 32;
lengthOfTerminator      =  1;
fieldNameOffset         = 16;  % Within table field descriptor
fieldNameLength         = 11;

% Get number of fields.
fseek(fid,8,'bof');
headerLength = fread(fid,1,'uint16');
numFields = (headerLength - lengthOfLeadingBlock - lengthOfTerminator)...
               / lengthOfDescriptorBlock;

% Read field lengths.
fseek(fid,lengthOfLeadingBlock + fieldNameOffset,'bof');
lengths = fread(fid,[1 numFields],'uint8',lengthOfDescriptorBlock - 1);

% Read the field names.
fseek(fid,lengthOfLeadingBlock,'bof');
data = fread(fid,[fieldNameLength numFields],...
             sprintf('%d*uint8=>char',fieldNameLength),...
             lengthOfDescriptorBlock - fieldNameLength);
data(data == 0) = ' '; % Replace nulls with blanks
names = cellstr(data')';

% Read field types.
fseek(fid,lengthOfLeadingBlock + fieldNameLength,'bof');
dbftypes = fread(fid,[numFields 1],'uint8=>char',lengthOfDescriptorBlock - 1);

% Convert DBF field types to MATLAB types.
typeConv = dbftype2matlab(upper(dbftypes));

% Return a struct array.
fieldInfo = cell2struct(...
    [names;  {typeConv.MATLABType}; {typeConv.ConvFunc}; num2cell(lengths)],...
    {'Name', 'Type',                'ConvFunc',          'Length'},1)';

%----------------------------------------------------------------------------
function typeConv = dbftype2matlab(dbftypes)
% Construct struct array with MATLAB types & conversion function handles.

typeLUT = ...
    {'N', 'double', @str2double2cell;...   % DBF numeric
     'F', 'double', @str2double2cell;...   % DBF float
     'C', 'char',   @cellstr;...           % DBF character
     'D', 'char',   @cellstr};             % DBF date

unsupported = struct('MATLABType', 'unsupported', ...
                     'ConvFunc',   @cellstr);

% Unsupported types: Logical,Memo,N/ANameVariable,Binary,General,Picture

numFields = length(dbftypes);
if numFields ~= 0
  typeConv(numFields) = struct('MATLABType',[],'ConvFunc',[]);
end
for k = 1:numFields
    idx = strmatch(dbftypes(k),typeLUT(:,1));
    if ~isempty(idx)
        typeConv(k).MATLABType = typeLUT{idx,2};
        typeConv(k).ConvFunc   = typeLUT{idx,3};
    else
        typeConv(k) = unsupported;
    end
end

%----------------------------------------------------------------------------
function out = str2double2cell(in)
% Translate IN, an M-by-N array of class char, to an M-by-1 column vector
% OUT, of class double.  IN may be blank- or null-padded. If IN(k,:) does
% not represent a valid scalar value, then OUT(k) has value NaN.
if isempty(in)
    out = {[NaN]};
    return
end

% Use sprintf when possible, but fall back to str2double for unusual cases.
fmt = sprintf('%%%df',size(in,2));
[data count] = sscanf(reshape(in',[1 numel(in)]),fmt);
if count == size(in,1)
    out = cell(count,1);
    for k = 1:count
        out{k} = data(k);
    end
else
    out = num2cell(str2double(cellstr(in)));
end

---

UPDATE

STR2DOUBLE2CELL subfunction sometimes works incorrectly if number of digits in the input parameter is different (see this discussion).

Here is my version of STR2DOUBLE2CELL:

function out = str2double2cell(in)
% Translate IN, an M-by-N array of class char, to an M-by-1 column vector
% OUT, of class double.  IN may be blank- or null-padded. If IN(k,:) does
% not represent a valid scalar value, then OUT(k) has value NaN.
if isempty(in)
    out = {[NaN]};
    return
end

out = cellfun(@str2double,cellstr(in),'UniformOutput',false);