Adaptive multi-dimensional data decomposition

A method of performing an Empirical Mode Decomposition on image or video data in two or more dimensions, the image or video data comprising an array of elements each having at least one value, the method comprising:

• determining (58) a local spatial average value for each element,
• wherein the local average at each element is defined as the mean (58) of two envelope functions at that element:
• a maximum envelope, comprising a function which interpolates among local maxima of the element values;
• and
• a minimum envelope, comprising a function which interpolates among local minima of the element values,
• the method further comprising generating (56a) the maximum envelope by:
• extracting (52) from the element values a sparse maximum signal which comprises the value of the element at each local maximum;
• and
• performing normalized convolution (100) of the sparse maximum signal with a smoothing filter;
• and
• generating (56b) the minimum envelope by:
• extracting (52) from the element values a sparse minimum signal which comprises the value of the element at each at local minimum;
• and
• performing normalized convolution (100) of the sparse minimum signal with a smoothing filter.

Signal processing apparatus adapted to perform an Empirical Mode Decomposition on numerical data, the data comprising an array of elements in two or more dimensions, each element comprising a value, the apparatus comprising:

• an averaging unit (58), adapted to determine for each element in the array a local average value,
• wherein the local average at each element is defined as the mean of two envelope functions at that element:
• a maximum envelope, comprising a function which interpolates among local maxima of the element values;
• and
• a minimum envelope, comprising a function which interpolates among local minima of the element values,
• the averaging unit further comprising:
• a local extrema extraction unit (52), adapted to extract from the element values a sparse maximum signal which comprises the element-value at each local maximum and a sparse minimum signal which comprises the element-value at each at local minimum;
• a maximum envelope generation unit (56a), adapted to generate the maximum envelope by performing normalized convolution of the sparse maximum signal with a smoothing filter;
• and
• a minimum envelope generation unit (56b), adapted generate the minimum envelope by performing normalized convolution of the sparse minimum signal with a smoothing filter.

Apparatus for decomposing an image or video into a plurality of components, the apparatus comprising:

• a luminance extraction unit (10), for obtaining an intensity signal of the image or video;
• and
• a signal decomposition unit (30) for decomposing the intensity signal into a component representing the three-dimensional structure of one or more objects in the scene and at least one other component,
• wherein the intensity signal is decomposed according to the spatial scale of variations in the intensity, and
• the other component comprises at least one of:
• a component representing surface reflectance of the objects;
• a component representing noise in the image;
• and
• a component representing illumination in the scene,
• the noise and reflectance components comprising variations at a small scale, the structure component comprising variations at a medium scale, and the illumination component comprising variations at a large scale or a uniform value.