tb_spectrogram — 2D typed op

• 数据种类:signal → image

• 调用:fullseye.apply(img, "tb_spectrogram", a=0.5, b=0.5)(2-D 的模型是一张图 + 两个标量旋钮 a,b∈[0,1])

用法

STFT 幅度谱图 -> `(freqs, times, S),其中 S 的形状为 (n_freqs, n_frames)。使用汉宁窗(Hann window);*hop* 默认值为 win//2`。

> 以下的详细说明为原文 —— 摘要与标题已翻译。

**Same raw convention as :func:spectrum, but a different divisor.** Each

column is the unnormalised `|rfft(frame * hann(win))|`, so it is not an

amplitude either — and dividing by `2/win` is *wrong* here, because the

Hann window has already thrown away part of the signal. The correct one-sided

amplitude conversion divides by the window's coherent gain::

w = np.hanning(win)

amp = S * (2.0 / w.sum()) # bins 1 .. win/2-1; DC / Nyquist: 1/w.sum()

Measured on a unit sine at a bin centre (`rate = 16000` Hz, 1000 Hz,

amplitude exactly 1.0, `win = 256`): the raw column peak is

`63.7497786196906; * 2/win gives 0.49804514546633283` (too small by

exactly the Hann coherent gain `sum(w)/win = 0.498046875`), while

`* 2/sum(w) gives 0.9999965273676957`. Only the second one is the

amplitude that was actually in the signal.

Peak *positions*, frame-to-frame ratios and any dB *difference* are unaffected

by either factor. This function returns magnitudes only — the phase is

discarded, so it cannot be inverted; use `acoustics.stft / acoustics.istft`

for a round-trip.

Typed bridge of the 1d op `spectrogram into the 2-D evolution registry: the same implementation, called under the op(v, a, b) convention. a drives rate (default 1) and b drives win` (default 256).

参考(示例数据・文献)

• 示例数据目录(下载 URL / 许可证) —— 2-D 用 skimage.data(BSD/公有领域)加合成图,3-D 给出真实数据源(Stanford/PDS 等)的下载 URL。

• 算子来历与参考文献 —— 该算子族所依据的研究/方法出处。

可运行的示例(实际调用该算子并已验证的样例)

• (暂无)

类型可衔接的下一个算子(可接受 image 作为输入)

identity · gaussian · mean_box · bilateral · unsharp · median · min_filter · max_filter

同类别(typed)

tb_points_to_voxel · tb_estimate_point_normals · tb_iss_keypoints · tb_angle_3points · tb_project_points · tb_render_point_depth · tb_statistical_outlier_removal · tb_radius_outlier_removal

*Provenance: ops.py — 2D 算子登记表。本条目由 tools/opdocs.py md 自动生成(请勿手工编辑)。*

© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.