dog

Documentation

Overview

package dog provides the Difference-of-Gaussians (DoG) filter for visual and other forms of signal processing

Index

• func GaussDenSig(x, sig float32) float32
• type Filter
  • func (gf *Filter) Defaults()
  • func (gf *Filter) FilterTensor(tsr *tensor.Float32, filt Filters) *tensor.Float32
  • func (gf *Filter) SetSize(sz, spc int)
  • func (gf *Filter) ShouldDisplay(field string) bool
  • func (gf *Filter) ToTable(tab *table.Table)
  • func (gf *Filter) ToTensor(tsr *tensor.Float32)
  • func (gf *Filter) Update()
• type Filters

Functions

func GaussDenSig

func GaussDenSig(x, sig float32) float32

GaussDenSig returns gaussian density for given value and sigma

Types

type Filter

type Filter struct {

	// is this filter active?
	On bool

	// how much relative weight does this filter have when combined with other filters
	Wt float32

	// overall gain multiplier applied after dog filtering -- only relevant if not using renormalization (otherwize it just gets renormed away)
	Gain float32 `default:"8"`

	// gain for the on component of filter, only relevant for color-opponent DoG's
	OnGain float32 `default:"1"`

	// size of the overall filter -- number of pixels wide and tall for a square matrix used to encode the filter -- filter is centered within this square -- typically an even number, min effective size ~6
	Size int

	// how far apart to space the centers of the dog filters -- 1 = every pixel, 2 = every other pixel, etc -- high-res should be 1 or 2, lower res can be increments therefrom
	Spacing int

	// gaussian sigma for the narrower On gaussian, in normalized units relative to Size
	OnSig float32 `default:"0.125"`

	// gaussian sigma for the wider Off gaussian, in normalized units relative to Size
	OffSig float32 `default:"0.25"`

	// cut off the filter (to zero) outside a circle of diameter = Size -- makes the filter more radially symmetric
	CircleEdge bool `default:"true"`
}

dog.Filter specifies a DoG Difference of Gaussians filter function.

func (*Filter) Defaults

func (gf *Filter) Defaults()

func (*Filter) FilterTensor

func (gf *Filter) FilterTensor(tsr *tensor.Float32, filt Filters) *tensor.Float32

FilterTensor extracts the given filter subspace from set of 3 filters in input tensor 0 = On, 1 = Off, 2 = Net

func (*Filter) SetSize

func (gf *Filter) SetSize(sz, spc int)

SetSize sets the size and spacing -- these are the main params that need to be varied for standard V1 dogs.

func (*Filter) ShouldDisplay

func (gf *Filter) ShouldDisplay(field string) bool

func (*Filter) ToTable

func (gf *Filter) ToTable(tab *table.Table)

ToTable renders filters into the given table.Table setting a column named Version and a column named Filter to the filter for that version (on, off, net) This is useful for display and validation purposes.

func (*Filter) ToTensor

func (gf *Filter) ToTensor(tsr *tensor.Float32)

ToTensor renders dog filters into the given table tensor.Tensor, setting dimensions to [3][Y][X] where Y = X = Size, and first one is On-filter, second is Off-filter, and third is Net On - Off

func (*Filter) Update

func (gf *Filter) Update()

type Filters

type Filters int

Filters is the type of filter

const (
	On Filters = iota
	Off
	Net
	FiltersN
)