Mean Filters

AI Summary

An AI-generated summary of this video lecture.

This lecture introduces spatial-domain image restoration for additive noise, focusing on mean filters. It begins with the degradation model g(x,y) = f(x,y) + η(x,y), where the original image is corrupted by unknown additive noise. Because the noise cannot be directly subtracted, spatial filtering is used to reduce it. The lecture categorizes de-noising filters into mean filters, order-statistics filters, and adaptive filters. The main focus is on four mean filters: arithmetic, geometric, harmonic, and contraharmonic. Each filter is defined by a formula and demonstrated on a 3×3 neighborhood containing a bright noisy pixel value of 200. The arithmetic mean replaces the center with approximately 65, the geometric mean with approximately 59, and the harmonic mean with 46. The contraharmonic filter is introduced with parameter Q, where positive Q removes pepper noise and negative Q removes salt noise. A comparison table summarizes the idea, formula, usefulness, image effect, and notes for each filter. The lecture concludes with X-ray circuit-board examples showing arithmetic, geometric, and contraharmonic filtering at Q = 15 and Q = -15.

Chapters

1. 0:00 – 2:00 00:00-02:00

   The lecture opens with a title slide reading “IMAGE RESTORATION” and the subtitle “Spatial Domain Restoration.” The instructor introduces restoration in the presence of noise, showing the degradation model g(x,y) = f(x,y) + η(x,y). A red box and underline are drawn around the equation to emphasize it. The slide explains that when additive noise is the only degradation, the degraded image consists of the original image plus unknown noise. Because the noise term is unknown and cannot be directly subtracted, spatial filtering is used to reduce it. A flowchart titled “SPATIAL FILTERS FOR DE-NOISING ADDITIVE NOISE” branches into three categories: MEAN FILTERS, ORDER-STATISTICS FILTERS, and ADAPTIVE FILTERS. The mean filters listed are Arithmetic, Geometric, Harmonic, and Contraharmonic Mean Filter; order-statistics filters include Median, Max, Min, and Midpoint.

2. 2:00 – 5:00 02:00-05:00

   The instructor continues explaining the spatial filtering approach, with red underlines highlighting phrases such as “the only degradation” and “unknown.” Red arrows point to the noise term in the formula g(x,y) = f(x,y) + η(x,y), reinforcing that the noise is additive and unknown. The flowchart under “SPATIAL FILTERS FOR DE-NOISING ADDITIVE NOISE” is revisited, and red checkmark arrows are drawn pointing down to each of the three filter category boxes: MEAN FILTERS, ORDER-STATISTICS FILTERS, and ADAPTIVE FILTERS. The slide emphasizes that spatial filters are mainly used to reduce additive noise, and the concept of replacing a noisy pixel alone is introduced. The transition leads into the detailed treatment of mean filters, with the four types listed: Arithmetic Mean Filter, Geometric Mean Filter, Harmonic Mean Filter, and Contraharmonic Mean Filter.

3. 5:00 – 10:00 05:00-10:00

   The lecture moves to the arithmetic mean filter, defined by the formula “Arithmetic Mean = Sum of pixel values / Number of pixels.” A 3×3 grid is shown with the center value “200” highlighted in red as a noisy pixel. The calculation box shows Sum = 588, Number of pixels = 9, and Arithmetic Mean = 588 / 9 ≈ 65.33 ≈ 65. A key point line reads “Arithmetic Mean → Normal Average → Noise Reduction → Possible Blurring.” Red hand-drawn circles and brackets are added around the grid and around the “9” and “65” in the calculation box. The slide then switches to the geometric mean filter, defined by “Geometric Mean = (Product of all pixel values)^(1/N).” A product line is shown ending in ≈ 59.15 ≈ 59, and the filtered center cell displays 59. Bottom rows on each slide show three small grids labeled “NOISY IMAGE (PART),” “APPLY ... FILTER,” and “FILTERED IMAGE (PART)” to illustrate the filtering steps.

4. 10:00 – 15:00 10:00-15:00

   The harmonic mean filter is introduced with the formula “Harmonic Mean = N / Σ(1/xi).” The same 3×3 neighborhood with center value “200” marked red as a very bright noisy pixel is used. Red hand-drawn annotations accumulate, including a box around the formula and a boxed “46” in the SOLUTION panel with an arrow. A bottom workflow reads “1. NOISY IMAGE (PART),” “2. APPLY HARMONIC MEAN FILTER,” and “FILTERED IMAGE (PART),” with the center replaced by 46. The lecture then transitions to the contraharmonic mean filter, listing “Q > 0 → Removes pepper noise (dark spots)” and “Q < 0 → Removes salt noise (bright spots).” This establishes the role of the parameter Q in controlling which type of impulse noise is reduced.

5. 15:00 – 17:55 15:00-17:55

   A comparison table titled “COMPARISON OF MEAN FILTERS” is displayed with four columns: Arithmetic, Geometric, Harmonic, and Contraharmonic Mean Filter. The rows are organized for Idea, How it works, Formula, Useful for, Effect on Image, and Note. The lecture then shows a slide headed “Example of all mean filters :” containing X-ray circuit-board images with captions referencing an arithmetic mean filter, a geometric mean filter, and contraharmonic filtering at Q = 15 and Q = -15. Red handwritten annotations appear at the top reading “+15,” “Q>0,” “Q<0,” and “-15” with checkmarks, plus red arrows pointing to specific image regions. The final frame returns to the initial slide with the equation g(x,y) = f(x,y) + η(x,y) and the flowchart under “SPATIAL FILTERS FOR DE-NOISING ADDITIVE NOISE,” closing the lecture on mean filters.

The lecture follows a clear pedagogical progression from the general noise model to specific filter implementations. It starts by establishing that additive noise is unknown and cannot be subtracted directly, motivating spatial filtering as the solution. The three broad categories of de-noising filters are introduced, with mean filters receiving detailed treatment. Each mean filter is presented with its formula, a worked example on a 3×3 neighborhood containing a bright noisy pixel of value 200, and the resulting filtered center value. The arithmetic mean yields approximately 65, the geometric mean approximately 59, and the harmonic mean 46, demonstrating that different means produce different levels of noise reduction. The contraharmonic filter is distinguished by its parameter Q, which controls whether pepper or salt noise is removed. The comparison table consolidates the four filters by idea, formula, usefulness, and effect on image quality. The final X-ray circuit-board examples provide practical visual evidence of each filter’s performance, with contraharmonic filtering shown at both Q = 15 and Q = -15 to illustrate the effect of sign choice. The lecture emphasizes that mean filters reduce additive noise but may introduce blurring, and that the choice of filter depends on the type of noise present.